CN117648873B - Ground subsidence prediction method, training method, device and equipment - Google Patents
Ground subsidence prediction method, training method, device and equipment Download PDFInfo
- Publication number
- CN117648873B CN117648873B CN202410121478.3A CN202410121478A CN117648873B CN 117648873 B CN117648873 B CN 117648873B CN 202410121478 A CN202410121478 A CN 202410121478A CN 117648873 B CN117648873 B CN 117648873B
- Authority
- CN
- China
- Prior art keywords
- observation data
- real
- prediction
- model
- ground subsidence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 91
- 238000012549 training Methods 0.000 title claims abstract description 85
- 238000012545 processing Methods 0.000 claims abstract description 50
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 27
- 238000001914 filtration Methods 0.000 claims abstract description 18
- 230000004044 response Effects 0.000 claims abstract description 12
- 239000011159 matrix material Substances 0.000 claims description 125
- 238000000605 extraction Methods 0.000 claims description 39
- 238000004364 calculation method Methods 0.000 claims description 23
- 238000003860 storage Methods 0.000 claims description 20
- 125000004122 cyclic group Chemical group 0.000 claims description 12
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 28
- 230000006870 function Effects 0.000 description 21
- 238000013528 artificial neural network Methods 0.000 description 20
- 238000004590 computer program Methods 0.000 description 18
- 230000008859 change Effects 0.000 description 16
- 239000003673 groundwater Substances 0.000 description 16
- 230000008569 process Effects 0.000 description 15
- 238000010586 diagram Methods 0.000 description 12
- 230000015654 memory Effects 0.000 description 10
- 238000009826 distribution Methods 0.000 description 8
- 239000013598 vector Substances 0.000 description 8
- 238000004891 communication Methods 0.000 description 7
- 238000003062 neural network model Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 230000006378 damage Effects 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 5
- 230000003287 optical effect Effects 0.000 description 5
- 238000009825 accumulation Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 238000010276 construction Methods 0.000 description 4
- 238000006073 displacement reaction Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 230000002123 temporal effect Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000035945 sensitivity Effects 0.000 description 3
- 239000002689 soil Substances 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000009933 burial Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005553 drilling Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 210000002364 input neuron Anatomy 0.000 description 2
- 238000005305 interferometry Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 210000004205 output neuron Anatomy 0.000 description 2
- 230000035699 permeability Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000004062 sedimentation Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000001010 compromised effect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000005489 elastic deformation Effects 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003090 exacerbative effect Effects 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 238000005755 formation reaction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 229910052500 inorganic mineral Inorganic materials 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 230000033001 locomotion Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000011707 mineral Substances 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 239000013535 sea water Substances 0.000 description 1
- 239000013049 sediment Substances 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a ground subsidence prediction method, a training method, a device and equipment, wherein the ground subsidence prediction method comprises the following steps: in response to received observation data aiming at a plurality of dimensions of a target area in a current period, synchronously processing the observation data to obtain real-time observation data; the method comprises the steps of utilizing a Kalman filtering algorithm to adjust model parameters of a pre-training model based on real-time observation data to obtain a target ground subsidence prediction model, wherein the pre-training model is obtained by training an initial model by utilizing historical observation data of a plurality of dimensions of a target area in a historical period; and inputting the real-time observation data into a target ground subsidence prediction model, and outputting prediction information, wherein the prediction information characterizes the probability of ground subsidence of a target area in a future period.
Description
Technical Field
The invention relates to the field of geological disaster prediction, in particular to a ground subsidence prediction method, a training device and ground subsidence prediction equipment.
Background
Ground deformation is an environmental geological phenomenon in which the elevation of the earth's surface changes continuously over a period of time. The ground subsidence is the main geological disaster in the ground deformation, and can bring serious harm and loss to human society, thereby threatening ecological safety. There is therefore a need for a timely and efficient method of predicting ground subsidence.
In the related art, a trained model is generally selected to predict ground subsidence, but more data sources influence ground subsidence prediction, and the data change of the data sources has randomness, and the trained model cannot be dynamically updated and adjusted to respond to the change condition of ground deformation, so that the accuracy of model prediction is difficult to improve.
Disclosure of Invention
In view of the above problems, the invention provides a ground subsidence prediction method, a training method, a device and equipment.
According to a first aspect of the present invention, there is provided a ground settlement prediction method comprising: in response to received observation data aiming at a plurality of dimensions of a target area in a current period, synchronously processing the observation data to obtain real-time observation data; the method comprises the steps of utilizing a Kalman filtering algorithm to adjust model parameters of a pre-training model based on real-time observation data to obtain a target ground subsidence prediction model, wherein the pre-training model is obtained by training an initial model by utilizing historical observation data of multiple dimensions of a target area in a historical period; and inputting the real-time observation data into the target ground subsidence prediction model, and outputting prediction information, wherein the prediction information characterizes the probability of ground subsidence of the target area in a future period.
According to an embodiment of the present invention, the foregoing adjusting model parameters of a pre-training model based on real-time observation data using a kalman filtering algorithm to obtain a target ground subsidence prediction model includes:
Acquiring a ground subsidence predicted value output by the pre-training model in a training stage and a label corresponding to the ground subsidence predicted value; obtaining a covariance matrix of ground subsidence prediction errors according to the ground subsidence prediction values and the labels corresponding to the ground subsidence prediction values; obtaining a Kalman gain matrix according to the covariance matrix of the ground settlement prediction error, the observed data matrix and the covariance matrix of the observed noise, wherein the observed data matrix is constructed according to the real-time observed data and the historical observed data; the covariance matrix of the observation noise is constructed according to the noise corresponding to the real-time observation data and the noise corresponding to the historical observation data; and adjusting the model parameters based on the real-time observation data by using the Kalman gain matrix to obtain a target ground subsidence prediction model.
According to an embodiment of the present invention, the real-time observation data includes S batches, S being an integer greater than 1; further comprises: obtaining a covariance matrix of a ground settlement prediction error corresponding to the real-time observation data of the s+1st batch according to a Kalman gain matrix corresponding to the real-time observation data of the S-1st batch, a covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the S-1st batch and an observation data matrix corresponding to the real-time observation data of the S-1st batch, wherein S is an integer greater than 1 and less than or equal to S-1; and obtaining a Kalman gain matrix corresponding to the real-time observation data of the s+1st batch according to the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s+1st batch, the observation data matrix corresponding to the real-time observation data of the s+1st batch and the covariance matrix of the observation noise corresponding to the real-time observation data of the s+1st batch.
According to an embodiment of the present invention, the synchronizing the observation data to obtain real-time observation data includes: according to a preset time length scale, performing time synchronization processing on the time-varying observation data in the observation data to obtain time-synchronized real-time observation data; and performing spatial synchronization processing on the observation data which changes with space in the observation data according to a preset spatial scale to obtain spatially synchronized real-time observation data.
According to the embodiment of the invention, the target ground subsidence prediction model comprises a convolution feature extraction module, a circulation feature extraction module and a prediction module; the inputting the real-time observation data into the target ground subsidence prediction model, outputting prediction information, including: extracting spatial features of the real-time observation data by using the convolution feature extraction module; processing the spatial features by using the cyclic feature extraction module to obtain time sequence features of the real-time observation data; and processing the time sequence characteristics by utilizing the prediction module to obtain the prediction information.
According to an embodiment of the present invention, further comprising: based on a Monte Carlo algorithm, constructing feature observation data according to the historical observation data of the multiple dimensions and the real-time observation data of the multiple dimensions; inputting the characteristic observation data into the target ground subsidence prediction model to obtain a prediction result corresponding to the characteristic observation data; analyzing the characteristic observation data and the prediction result to obtain a target observation data type for representing the influence on the prediction result and an early warning critical value corresponding to the target observation data type; and generating early warning information for representing that the ground subsidence occurs in the target area in a future period in response to the value of the real-time observed data corresponding to the target observed data type being greater than the early warning critical value.
According to a second aspect of the present invention, there is provided a model training method comprising: acquiring sample historical observation data of a plurality of dimensions of a sample area in a historical period; synchronizing the sample historical observation data to obtain target sample historical observation data; inputting the historical observation data of the target sample into an initial model to obtain sample prediction information, wherein the sample prediction information characterizes the probability of ground subsidence of the sample area in a target period; obtaining a loss value based on the loss function according to the sample prediction information and the sample label; and adjusting model parameters of the initial model based on the loss value to obtain a pre-training model, wherein the pre-training model is applied to the ground subsidence prediction method.
A third aspect of the present invention provides a ground subsidence prediction apparatus comprising: the system comprises a synchronization module, an adjustment module and a prediction module.
The synchronization module is used for responding to the received observation data of multiple dimensions of the target area in the current period, and performing synchronization processing on the observation data to obtain real-time observation data; the adjustment module is used for adjusting model parameters of a pre-training model based on real-time observation data by utilizing a Kalman filtering algorithm to obtain a target ground subsidence prediction model, wherein the pre-training model is obtained by training an initial model by utilizing historical observation data of a plurality of dimensions of the target area in a historical period; and the prediction module is used for inputting the real-time observation data into the target ground subsidence prediction model and outputting prediction information, wherein the prediction information represents whether ground subsidence occurs in the target area in a future period.
According to an embodiment of the invention, the adjustment module comprises: the device comprises a label acquisition sub-module, a covariance calculation sub-module, a Kalman gain matrix calculation sub-module and a parameter adjustment sub-module. The label obtaining sub-module is used for obtaining the ground subsidence predicted value output by the pre-training model in the training stage and the label corresponding to the ground subsidence predicted value; the covariance calculation sub-module is used for obtaining a covariance matrix of the ground subsidence prediction error according to the ground subsidence prediction value and the label corresponding to the ground subsidence prediction value; the Kalman gain matrix calculation sub-module is used for obtaining a Kalman gain matrix according to the covariance matrix of the ground subsidence prediction error, the covariance matrix of the observation data matrix and the observation noise, wherein the observation data matrix is constructed according to the real-time observation data and the historical observation data; the covariance matrix of the observation noise is constructed according to the noise corresponding to the real-time observation data and the noise corresponding to the historical observation data; and a parameter adjustment sub-module, configured to adjust the model parameter based on the real-time observation data by using the kalman gain matrix, so as to obtain a target ground subsidence prediction model.
According to an embodiment of the present invention, the real-time observation data includes S batches, S is an integer greater than 1, and the adjustment module further includes: a first covariance calculation sub-module and a first Kalman gain matrix calculation sub-module. The first covariance calculation sub-module is used for obtaining the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s+1st batch according to the Kalman gain matrix corresponding to the real-time observation data of the S-1st batch, the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the S-1st batch and the observation data matrix corresponding to the real-time observation data of the S-1st batch, wherein S is an integer greater than 1 and less than or equal to S-1. The first Kalman gain matrix calculation sub-module is used for obtaining a Kalman gain matrix corresponding to the real-time observation data of the s+1st batch according to the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s+1st batch, the observation data matrix corresponding to the real-time observation data of the s+1st batch and the covariance matrix of the observation noise corresponding to the real-time observation data of the s+1st batch.
According to an embodiment of the invention, the synchronization module comprises: a temporal synchronization sub-module and a spatial synchronization sub-module. The time synchronization sub-module is used for carrying out time synchronization processing on the observation data which changes along with time in the observation data according to a preset time length scale to obtain time-synchronized real-time observation data; and the space synchronization sub-module is used for carrying out space synchronization processing on the observation data which changes with space in the observation data according to a preset space scale to obtain space synchronization real-time observation data.
According to the embodiment of the invention, the target ground subsidence prediction model comprises a convolution feature extraction module, a circulation feature extraction module and a prediction module; the prediction module comprises a convolution extraction sub-module, a circulation extraction sub-module and an information prediction sub-module. The convolution extraction submodule is used for extracting the spatial characteristics of the real-time observation data by utilizing the convolution characteristic extraction module; the cyclic extraction sub-module is used for processing the spatial characteristics by utilizing the cyclic characteristic extraction module to obtain the time sequence characteristics of the real-time observation data; and the prediction sub-module is used for processing the time sequence characteristics by utilizing the prediction module to obtain the prediction information.
According to an embodiment of the present invention, the ground subsidence prediction apparatus further includes: the system comprises a feature observation data construction module, a result prediction module, a prediction result analysis module and an early warning information generation module.
The characteristic observation data construction module is used for constructing characteristic observation data according to the historical observation data of the plurality of dimensions and the real-time observation data of the plurality of dimensions based on a Monte Carlo algorithm; the result prediction module is used for inputting the characteristic observation data into the target ground subsidence prediction model to obtain a prediction result corresponding to the characteristic observation data; the prediction result analysis module is used for analyzing the characteristic observation data and the prediction result to obtain a target observation data type for representing the influence on the prediction result and an early warning critical value corresponding to the target observation data type; and the early warning information generation module is used for generating early warning information used for representing that the ground subsidence occurs in the target area in a future period in response to the value of the real-time observed data corresponding to the target observed data type is larger than the early warning critical value.
A fourth aspect of the present invention provides a model training apparatus comprising: the system comprises an acquisition module, a data synchronization module, a processing module, a loss module and a pre-training module. The acquisition module is used for acquiring sample historical observation data of a plurality of dimensions of the sample area in a historical period; the data synchronization module is used for performing synchronization processing on the sample historical observation data to obtain target sample historical observation data; the processing module is used for inputting the historical observation data of the target sample into an initial model to obtain sample prediction information, wherein the sample prediction information characterizes the probability of ground subsidence of the sample area in a target period; the loss module is used for obtaining a loss value based on the loss function according to the sample prediction information and the sample label; and the pre-training module is used for adjusting the model parameters of the initial model based on the loss value to obtain a pre-training model, wherein the pre-training model is applied to the ground subsidence prediction method.
A fifth aspect of the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the above-described ground settlement prediction method.
A fourth aspect of the invention also provides a computer readable storage medium having stored thereon executable instructions which, when executed by a processor, cause the processor to perform the above-described ground settlement prediction method.
The fifth aspect of the present invention also provides a computer program product comprising a computer program which when executed by a processor implements the above ground settlement prediction method.
According to the invention, real-time observation data is obtained by processing the observation data of multiple dimensions in the current period, and parameters of the pre-training model are adjusted based on the real-time observation data by utilizing a Kalman filtering algorithm, so that the ground subsidence is predicted through the pre-training model. The pre-training model can be optimized according to the real-time observation result, the influence of the randomness of the observation data on the model precision is reduced, and therefore the accuracy of model prediction is further improved.
Drawings
The foregoing and other objects, features and advantages of the invention will be apparent from the following description of embodiments of the invention with reference to the accompanying drawings, in which:
FIG. 1 shows an application scenario diagram of a ground settlement prediction method according to an embodiment of the invention;
FIG. 2 shows a flow chart of a ground settlement prediction method according to an embodiment of the invention;
FIG. 3 shows a flow chart of a model training method according to an embodiment of the invention;
FIG. 4 shows a flow chart of deriving predictive information in accordance with an embodiment of the invention;
FIG. 5 shows a flow chart of a synchronization process for observed data according to an embodiment of the present invention;
FIG. 6 shows a flow chart for deriving a target ground subsidence prediction model in accordance with an embodiment of the present invention;
FIG. 7 shows a flow chart for targeted generation of early warning information of ground subsidence in accordance with an embodiment of the present invention;
FIG. 8 shows a block diagram of a ground settlement prediction apparatus according to an embodiment of the invention;
FIG. 9 shows a block diagram of a model training apparatus according to an embodiment of the present invention; and
Fig. 10 shows a block diagram of an electronic device adapted to implement a ground settlement prediction method according to an embodiment of the invention.
Detailed Description
Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. It should be understood that the description is only illustrative and is not intended to limit the scope of the invention. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the invention. It may be evident, however, that one or more embodiments may be practiced without these specific details. In addition, in the following description, descriptions of well-known structures and techniques are omitted so as not to unnecessarily obscure the present invention.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The terms "comprises," "comprising," and/or the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components.
All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It should be noted that the terms used herein should be construed to have meanings consistent with the context of the present specification and should not be construed in an idealized or overly formal manner.
Where a convention analogous to "at least one of A, B and C, etc." is used, in general such a convention should be interpreted in accordance with the meaning of one of skill in the art having generally understood the convention (e.g., "a system having at least one of A, B and C" would include, but not be limited to, systems having a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.).
Ground deformation refers to an environmental geological phenomenon in which the elevation of the earth's surface changes continuously over a period of time, including ground subsidence and rebound. Wherein, the ground subsidence is the main geological hazard in the current ground deformation, and the ground subsidence has natural ground subsidence and artificial ground subsidence. Natural ground subsidence is one of the loose or semi-loose sedimentary formations from loose to fine diagenetic processes under the action of gravity and the other is due to geologic structure movement, earthquakes, etc. The artificial ground subsidence mainly has three reasons, namely, the first is to develop and utilize underground fluid resources, the second is karst collapse and the third is to mine solid mineral products.
The ground subsidence can bring serious harm and loss to human society, such as building inclination, cracks, collapse, foundation destruction, pipeline rupture, road cracks, bridge rupture and the like; also can affect the hydrologic environment, such as changing hydrologic cycle, causing flood disasters, exacerbating seawater invasion, etc.; ecological safety is also compromised, such as soil loss, vegetation degradation, reduced biodiversity, etc. Therefore, the ground settlement situation can be effectively monitored and early-warned in time, and the method has important significance for preventing and reducing the harm of the ground settlement situation.
The methods commonly used at present are as follows:
Ground deformation monitoring instruments, such as level, GNSS (Global Navigation SATELLITE SYSTEM ) receiver, optical measuring instrument, deep displacement instrument, etc., are used to measure the relevant parameters of ground deformation, such as elevation, displacement, speed, acceleration, etc., periodically or in real time by installing monitoring devices on the ground or underground. The method has higher precision and reliability, but also has some disadvantages such as high cost, small coverage range, easy artificial interference and the like.
The remote sensing monitoring method, such as synthetic aperture radar interferometry (InSAR), optical remote sensing method, etc., uses satellite or aircraft carried remote sensing sensor to periodically or real-time obtain the related image of ground deformation, and uses image processing technology to extract the related information of ground deformation, such as elevation change, displacement change, etc. The method has the advantages of large coverage, low cost, no need of manual intervention and the like, but also has the disadvantages of low precision, easiness in atmospheric interference, multi-phase image requirement and the like.
Numerical simulation methods, such as finite element methods, finite difference methods, boundary element methods and the like, are used for carrying out numerical solution by a computer according to known initial conditions and boundary conditions by establishing a mathematical model of ground deformation, so as to obtain related parameters of the ground deformation, such as elevation change, displacement change and the like. The method has the advantages of simulating complex conditions, predicting future changes and the like, but also has the disadvantages of large input data, large calculation amount, error accumulation and the like.
The method has advantages and disadvantages, but cannot meet the requirements of real-time, accurate and comprehensive monitoring and early warning of ground deformation.
Accordingly, an embodiment of the present invention provides a ground subsidence prediction method including: in response to received observation data aiming at a plurality of dimensions of a target area in a current period, synchronously processing the observation data to obtain real-time observation data; the method comprises the steps of utilizing a Kalman filtering algorithm to adjust model parameters of a pre-training model based on real-time observation data to obtain a target ground subsidence prediction model, wherein the pre-training model is obtained by training an initial model by utilizing historical observation data of a plurality of dimensions of a target area in a historical period; and inputting the real-time observation data into a target ground subsidence prediction model, and outputting prediction information, wherein the prediction information characterizes the probability of ground subsidence of a target area in a future period.
Fig. 1 shows an application scenario diagram of a ground subsidence prediction method according to an embodiment of the present invention.
As shown in fig. 1, an application scenario 100 according to this embodiment may include a first data acquisition unit 101, a second data acquisition unit 102, a third data acquisition unit 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the first data acquisition unit 101, the second data acquisition unit 102, the third data acquisition unit 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.
The first data acquisition unit 101, the second data acquisition unit 102 and the third data acquisition unit 103 may include a leveling data acquisition unit for acquiring ground elevation data measured by a leveling instrument; may include a GPS (Global Positioning System ) data acquisition unit for acquiring ground coordinate data measured by a GPS receiver; the system can comprise an InSAR (Interferometric Synthetic Aperture Radar, synthetic aperture radar interference) data acquisition unit, which is used for acquiring ground deformation interference image data shot by a synthetic aperture radar satellite or an airplane; the system can comprise a groundwater level data acquisition unit for acquiring groundwater level information; a borehole data acquisition unit may also be included for acquiring sand-stick thickness of the borehole and total borehole thickness, wherein sand-stick thickness refers to the thickness of sandy soil and the thickness of stick soil.
The first data acquisition unit 101, the second data acquisition unit 102, the third data acquisition unit 103 interact with the server 105 via the network 104 to receive or send messages or the like. The data acquisition unit may consist of different kinds of data acquisition as mentioned above.
The server 105 may be a server providing various services. The background management server can analyze and process the received data such as the request and the like, and feed back the processing result to the data acquisition unit.
It should be noted that the ground subsidence prediction method according to the embodiment of the present invention may be generally performed by the server 105. Accordingly, the ground subsidence prediction apparatus according to the embodiment of the present invention may be generally disposed in the server 105. The ground settlement prediction method provided by the embodiment of the invention may also be performed by a server or a server cluster which is different from the server 105 and is capable of communicating with the first data acquisition unit 101, the second data acquisition unit 102, the third data acquisition unit 103 and/or the server 105. Accordingly, the ground settlement prediction device provided by the embodiment of the invention may also be arranged in a server or a server cluster which is different from the server 105 and is capable of communicating with the first data acquisition unit 101, the second data acquisition unit 102, the third data acquisition unit 103 and/or the server 105.
It should be understood that the number of data acquisition units, networks and servers in fig. 1 is merely illustrative. There may be any number of data acquisition units, networks, and servers, as desired for implementation.
Fig. 2 shows a flow chart of a ground settlement prediction method according to an embodiment of the invention.
As shown in fig. 2, the ground subsidence prediction in this embodiment includes operations S210 to S230.
In operation S210, in response to the received observation data for the plurality of dimensions of the target area within the current period, the observation data is synchronously processed, resulting in real-time observation data.
According to the embodiment of the invention, the observation data with multiple dimensions is a multi-dimensional data set with a unified format, and for single data, the observation data can comprise deformation rate, deformation time sequence, water level elevation, water level change rate, water level time sequence, lithology, water group distribution, groundwater exploitation amount and other attributes. The above-mentioned attribute can be extracted from satellite image to ground deformation information by InSAR, including deformation rate, deformation accumulation, deformation time sequence, etc.; monitoring points can be set on the ground by using instruments such as a level gauge, a GPS (global positioning system), a fiber bragg grating and the like, and ground deformation information including deformation rate, deformation accumulation amount, deformation time sequence and the like can be measured periodically or in real time; the underground water level change at the underground water burial depth can be measured by using facilities such as a water level gauge, a hydrological well and the like, and the facilities comprise water level elevation, water level change rate, water level time sequence and the like; the underground water level change at the underground water burial depth can be measured by using facilities such as a water level gauge, a hydrological well and the like, and the facilities comprise water level elevation, water level change rate, water level time sequence and the like; and the underground water exploitation amount can be obtained by using statistical data and the like.
For ease of understanding, some concepts related to embodiments of the invention are explained:
Deformation rate: represents the distance of deformation of the ground per unit time, in mm/year.
Deformation cumulative amount: the total distance of the ground deformation from a certain reference point in time to the current point in millimeters is indicated.
Deformation time sequence: the cumulative amount of deformation in millimeters at each time point over a period of time is expressed.
Elevation of water level: the height of the groundwater level at the current point in meters with respect to the sea level is indicated.
Rate of change of water level: the distance of the change of the groundwater level in unit time is expressed in meters per year.
Lithology: indicating the type of deposit at the current location, such as sand, sticking, etc.
Aqueous group distribution: indicating whether the current location has an aqueous group and the depth range of the aqueous group in meters.
Groundwater exploitation amount: represents the volume of groundwater produced per year per unit area of the current location in cubic meters per square kilometer per year.
According to the embodiment of the invention, the observed data can be preprocessed to obtain the observed data with higher quality.
According to the embodiment of the invention, the collected data can be converted into a unified format, for example, the collected data can be uniformly stored into CSV, TXT format and the like, so that subsequent processing and storage are convenient.
According to the embodiment of the invention, the operation such as dimensionless normalization processing can be performed on the observed data, so that the numerical range is between 0 and 1, the dimensionality difference of the data is reduced, and the convergence speed and stability of the model are improved.
The data may be normalized by the following formula (1):
（1）
Wherein x i is the observed point measured sample value, x min is the minimum sample value, x max is the maximum sample value, and x * is the normalized data value.
In operation S220, model parameters of the pre-training model are adjusted based on real-time observation data using a kalman filtering algorithm to obtain a target ground subsidence prediction model.
The pre-training model is obtained by training an initial model by utilizing historical observation data of a target area in multiple dimensions in a historical period. For the pre-trained model, due to the fact that the real-time observation data and the historical observation data have certain differences, the Kalman filtering algorithm can be utilized to adjust model parameters of the pre-trained model based on the real-time observation data so as to obtain optimal model parameters matched with the real-time observation data, and prediction accuracy can be effectively improved.
In operation S230, the real-time observation data is input into the target ground subsidence prediction model, and the prediction information is output.
Wherein the predictive information characterizes a probability of ground subsidence of the target area occurring at a future time period.
According to the embodiment of the invention, after the optimized model parameters are obtained, the pre-trained model with the optimized model parameters is used as a target ground subsidence prediction model, and real-time observation data is input into the target ground subsidence prediction model to obtain prediction information output by the model.
According to the invention, real-time observation data is obtained by processing the observation data of multiple dimensions in the current period, and parameters of the pre-training model are adjusted based on the real-time observation data by utilizing a Kalman filtering algorithm, so that the ground subsidence is predicted through the pre-training model. The pre-training model can be optimized according to the real-time observation result, the influence of the randomness of the observation data on the model precision is reduced, and therefore the accuracy of model prediction is further improved.
Since the input layer of the model of the observation data with multiple dimensions is provided with multiple nodes, the model corresponds to the observation data with different dimensions respectively. The number of nodes of the input layer can influence the structure and the parameter number of the neural network, thereby influencing the learning capacity and the efficiency of the neural network, and even solving the problems of parameter gradient disappearance or gradient explosion and the like.
Therefore, the invention also provides a model training method, which introduces a Kalman filtering algorithm to process data so as to solve the problems of the elimination of the parameter gradient and explosion.
Based on the above ground subsidence prediction method, fig. 3 of the present invention shows a flowchart of a model training method according to an embodiment of the present invention.
As shown in FIG. 3, the model training method of this embodiment includes operations S310-S350.
In operation S310, sample history observation data of a plurality of dimensions of a sample region within a history period is acquired.
According to an exemplary embodiment of the present invention, quality detection may be performed on the collected data, for example, an InSAR (Interferometric Synthetic Aperture Radar, synthetic aperture radar interferometry) inversion result may be calibrated and verified with accuracy by using the signpost observation data, the actually measured level point and the GPS (Global Positioning System ) data, for example, a contrast correction model may be established by using a linear fitting method according to the correlation between the leveling data and the GPS data on the single point and the InSAR data, and the data comparison and correction may be performed on a large range of InSAR data by using the model to generate a ground subsidence data set.
The collected data can be subjected to abnormal value detection, noise filtering, missing value filling and other operations so as to improve the quality of the observed data. And the method can also judge whether the deformation of the sediment is elastic deformation or plastic deformation according to the comparison between the current water level value and the earlier lowest water level value, and transmit the information to a machine learning training model.
According to the embodiment of the invention, the correlation degree between the observed data and the ground subsidence can be analyzed, and important influencing factors are selected as a model data set. The Pearson linear correlation can be used for judging the autocorrelation degree among the potential main control factors (groundwater level, mining amount, hydrogeological parameters, lithology and the like of each aquifer), eliminating the main control factors of potential autocorrelation (determining coefficient R 2 > 0.8), and taking the main control factors of non-autocorrelation as a vector set of machine learning influence factors.
In operation S320, the sample history observation data is subjected to synchronization processing to obtain target sample history observation data.
In operation S330, the target sample history observation data is input into the initial model to obtain sample prediction information.
Wherein the sample prediction information characterizes a probability of ground subsidence of the sample region at the target time period.
According to the embodiment of the invention, the historical observation data of the target area in multiple dimensions in the historical period can be divided into a training set and a testing set, wherein the training set accounts for 80% and the testing set accounts for 20%. The invention adopts the deep neural network to train the pre-training model, has stronger nonlinear fitting capability and generalization capability, can effectively utilize the multisource characteristics of the observed data, and improves the accuracy and the instantaneity of ground subsidence prediction.
Deep neural networks are nonlinear function approximators composed of multiple layers of neurons, each of which can perform some transformation or abstraction on the input data, thus achieving layer-by-layer extraction and representation from low-level features to high-level features. The structure of the deep neural network can be flexibly designed and adjusted according to different tasks and data, and common structure types include a full-connection network, a convolutional neural network, a cyclic neural network and the like.
According to the embodiment of the invention, at the beginning of model training, the model can be initialized first, and all parameters (namely weights and biases) of the deep neural network model are initialized randomly so as to be subjected to certain distribution, such as uniform distribution or normal distribution. And selecting proper super parameters such as learning rate, batch size, iteration times and the like, and controlling the training process and effect of the model.
And secondly, inputting data in the training set into a deep neural network model, and obtaining an output result of the model, namely the early warning probability, through calculation of a part of the convolution feature extraction module, the circulation feature extraction module and the prediction module.
In operation S340, a loss value is obtained from the sample prediction information and the sample tag based on the loss function.
According to the embodiment of the invention, the loss function of the model can be calculated according to the output result of the model and the real early warning label, and the prediction error of the model can be reflected. The present invention selects the root mean square error between the predicted ground subsidence and the actual observed ground subsidence as the loss function.
In operation S350, model parameters of the initial model are adjusted based on the loss values, resulting in a pre-trained model.
According to the embodiment of the invention, gradient calculation is carried out on all parameters of the deep neural network model according to the loss value, and the gradient of the parameters is sequentially solved from an output layer to an input layer by utilizing a chain rule. According to the gradient descent rule, all parameters of the model are updated to reduce the loss function value of the model.
The performance of the model after data processing can be evaluated by utilizing the decision coefficient (R 2) of the predicted ground subsidence value and the observed ground subsidence value, and model evaluation indexes such as accuracy, recall rate and the like are calculated to reflect the early warning performance of the model. And after each iteration period is finished, comparing the evaluation index of the current model with the best evaluation index before, if the current model is better, storing the parameters of the current model, and updating the best evaluation index. Otherwise, the original optimal model is kept unchanged. And judging whether the preset iteration times are reached or whether the loss function value is converged to a certain degree, if so, stopping the training process, and outputting the optimal model and the evaluation index thereof. And otherwise, continuing iteration.
According to an embodiment of the invention, a set Kalman filtering algorithm can be used in combination with a deep neural network for processing, which is a method for estimating uncertainty of a system state and updating the system state by using a set of samples of model states. It is a variant of the kalman filter algorithm, suitable for non-linear, high-dimensional systems.
The ensemble kalman filter algorithm can be divided into the following steps:
An initial set of parameters is generated. Assuming that the neural network parameters are assumed to satisfy the normal distribution, randomly extracting N samples from the initial normal distribution as an initial set of parameters, and recording as Wherein x refers to a state parameter, and i refers to a neural network connection position.
A sample prediction information set. Inputting the initial set of parameters into the model by utilizing the neural network model, removing the parameter set with larger fitting error (root mean square error) according to the comparison of the output result and the ground subsidence observation value, reserving the parameter set with better fitting effect as a sample prediction information set, and marking asS represents a batch. The mean value of the sample prediction information set may be calculated according to the following formula (2).
（2）
Wherein,Representing the mean of a sample prediction information set,/>Sample prediction information of lot s indicating the i-th neural network connection position, and N indicates the number of sample prediction information in the sample prediction information set.
The covariance of the sample prediction information set may be calculated according to the following equation (3).
（3）
Wherein,Representing covariance of sample prediction information set,/>Sample prediction information of lot s representing ith neural network connection location,/>The average value of the sample prediction information set is represented, and N represents the number of sample prediction information in the sample prediction information set.
Since operation S230 is similar to the operation of obtaining the pre-training model, how the prediction information is obtained in operation S230 will be exemplarily described as follows through fig. 4.
As shown in FIG. 4, the process of obtaining the target ground subsidence prediction model may include operations S410-S430.
The target ground subsidence prediction model comprises a convolution feature extraction module, a circulation feature extraction module and a prediction module.
In operation S410, spatial features of real-time observation data are extracted using a convolution feature extraction module.
According to the embodiment of the invention, the convolution feature extraction module is mainly used for extracting spatial features from the data set after the data fusion, and comprises a plurality of convolution layers and a pooling layer. The convolution layer is a layer for filtering and feature mapping the input data by using a local receptive field and a weight sharing mode. The pooling layer is a layer that downsamples and invariance enhances the input data. By means of the convolution feature extraction section, the input data can be converted into spatial feature vectors having higher dimensions and more abstract meanings.
In operation S420, the spatial features are processed by the cyclic feature extraction module to obtain timing features of the real-time observation data.
According to the embodiment of the invention, the cyclic feature extraction module is mainly used for extracting time sequence features from the spatial feature vectors output by the convolution feature extraction part, and comprises an LSTM (Long short-term memory network). The LSTM is a circulating neural network capable of solving the long-term dependence problem, has three gating structures of an input gate, a forgetting gate and an output gate, and can effectively control the storage and forgetting of information. By the cyclic feature extraction section, the spatial feature vector can be converted into a temporal feature vector having a higher dimension and a more abstract meaning.
In operation S430, the temporal feature is processed using the prediction module to obtain prediction information.
According to the embodiment of the invention, the prediction module is mainly used for generating a ground deformation early warning result according to the time sequence feature vector output by the cyclic feature extraction part, and the ground deformation early warning result comprises a full-connection layer and an activation function. The full connection layer is a layer connecting all input neurons with all output neurons, and can realize linear transformation of input data. The activation function is a function of performing nonlinear transformation on input data, and can enhance the expressive power of the network.
According to the embodiment of the invention, in order to realize the classification problem (namely early warning or no early warning), a sigmoid function is selected as an activation function, and the output value of the sigmoid function is between 0 and 1 and can be expressed as the early warning probability.
According to the embodiment of the invention, the convolution feature extraction module is used for extracting the space features of the real-time observation data and the time sequence features of the real-time observation data are obtained through processing of the circulation feature extraction module to predict the ground subsidence, so that the observation data with multiple dimensions can be fully utilized to predict the ground subsidence.
How the observed data is synchronously processed in operation S210 will be exemplarily described by fig. 5 as follows.
As shown in FIG. 5, operations S510-S520 may be included in the flow of synchronizing the observation data.
In operation S510, time synchronization processing is performed on time-varying observation data in the observation data according to a predetermined time scale, so as to obtain time-synchronized real-time observation data.
According to the embodiment of the invention, the collected observation data can be time-synchronized, for example, the collected observation data can be unified into the month scale vector data so as to carry out subsequent processing. The daily scale data comprises leveling data, GPS data, inSAR data, groundwater level data and the like, and a month average value can be calculated on a month scale; annual scale data, including mining volume and the like, can be distributed to month scales according to historical records by utilizing historical data; time stable data, comprising: lithology data can be made into vector data on a month scale according to constants.
In operation S520, spatial synchronization processing is performed on the observation data that varies spatially among the observation data according to a predetermined spatial scale, to obtain spatially synchronized real-time observation data.
According to an embodiment of the present invention, spatial data statistics are distributed to a spatial grid (e.g., 5 km) in order to generate enough data samples (including ground subsidence, groundwater level, production, lithology, etc.) for machine learning. Interpolation is carried out on data (such as the exploitation quantity, the groundwater level and the like) lower than the space grid precision by utilizing a Kriging interpolation method to obtain corresponding grid data; for data above this spatial grid accuracy (e.g., inSAR data), a weighted average method may be used to obtain an average over the grid space.
According to the embodiment of the invention, the multidimensional data with uniform dimensions are obtained by synchronizing the time dimension and the space dimension of the multidimensional observation data, and are input into the model for the next processing.
How the model parameters of the pre-trained model are adjusted to obtain the target ground subsidence prediction model using the kalman filter algorithm in operation S220 will be exemplarily described as follows through fig. 6.
As shown in FIG. 6, the process of obtaining the target ground subsidence prediction model may include operations S610-S640. In operation S610, a ground subsidence prediction value output by the pre-training model in the training phase and a tag corresponding to the ground subsidence prediction value are acquired.
In operation S620, a covariance matrix of the ground settlement prediction error is obtained according to the ground settlement prediction value and the label corresponding to the ground settlement prediction value.
According to an embodiment of the present invention, the real-time observation data may include S batches, S being an integer greater than 1. For the real-time observation data of the 1 st batch, a covariance matrix of the ground subsidence prediction error can be obtained according to the ground subsidence prediction value and the label corresponding to the ground subsidence prediction value.
In operation S630, a kalman gain matrix is obtained from the covariance matrix of the ground settlement prediction error, the observed data matrix, and the covariance matrix of the observed noise.
The observation data matrix is constructed according to real-time observation data and historical observation data; the covariance matrix of the observation noise is constructed from the noise corresponding to the real-time observation data and the noise corresponding to the historical observation data.
According to the embodiment of the present invention, the observed data matrix and the observed noise may be calculated similarly to the above step S620 to obtain the covariance matrix of the observed data matrix and the observed noise. Furthermore, the Kalman gain matrix can be obtained according to the covariance matrix of the ground subsidence prediction error, the covariance matrix of the observed data matrix and the observed noise.
For the real-time observation data of the S-th batch, S is an integer of more than 1 and less than or equal to S-1. The following operations may be performed: and obtaining the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s+1st batch according to the Kalman gain matrix corresponding to the real-time observation data of the s batch, the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s-1 st batch and the observation data matrix corresponding to the real-time observation data of the s batch.
According to the embodiment of the invention, a Kalman gain matrix corresponding to the real-time observation data of the s+1st batch is obtained according to the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s+1st batch, the observation data matrix corresponding to the real-time observation data of the s+1st batch and the covariance matrix of the observation noise corresponding to the real-time observation data of the s+1st batch.
According to the embodiment of the invention, after receiving the observation data, a Kalman gain matrix is established to correct the model parameters in the pre-training model, for example, the optimal ground subsidence estimation and error covariance estimation can be obtained according to the predicted ground subsidence and error covariance.
The kalman gain matrix can be calculated by the following formula (4).
（4）
Wherein K s+1 is a kalman gain matrix corresponding to the observation data of the s+1st lot, which reflects the specific gravity of the prediction error and the observation error, and the larger K s+1 is, the larger the influence of the observation data on the ground subsidence estimation is, and the smaller the influence is on the ground subsidence estimation is. P s is an error covariance optimum parameter estimate corresponding to the s-th batch of observed data, which may also be referred to as a posterior error covariance estimate, H is a ground settlement observation matrix, and R is a covariance matrix of ground settlement observation noise.
In operation S640, model parameters are adjusted based on the real-time observation data using the kalman gain matrix, to obtain a target ground subsidence prediction model.
According to the embodiment of the invention, the predicted ground subsidence and the error covariance can be corrected through the following formula (5) and formula (6), so that the optimal parameter estimation and the optimal error covariance estimation are obtained.
（5）
（6）
Wherein,And/>The optimal estimates of the parameters for the s+1 and s batches, z s+1 is the latest ground settlement observation data, P s+1 and P s are the optimal estimates of the error covariance for the s+1 and s batches, respectively, and I is the identity matrix.
According to the embodiment of the invention, the Kalman gain matrix is constructed by calculating the covariance matrix of the ground subsidence prediction error and the covariance matrix of the observation noise in the real-time observation data, and the model parameters are adjusted in real time by utilizing the Kalman gain matrix, so that the influence of the noise in the real-time observation data on the real-time observation data can be effectively removed, and the prediction result of the model is more accurate.
According to the ground subsidence prediction method, the prediction information of the ground subsidence probability of the target area in the future period can be obtained according to the real-time observation information, and in order to further test the sensitivity of the prediction data in the data class in the real-time observation information, the observation data affecting the ground subsidence is further analyzed based on the Monte Carlo algorithm so as to obtain the sensitive data in the observation data and the corresponding sensitive threshold value, so that the ground subsidence prediction can be better carried out according to the sensitive data threshold value, and the operation pressure of equipment is reduced.
Because the factors influencing the ground subsidence are numerous, in order to further improve the timeliness of the ground subsidence early warning, the multidimensional data can be analyzed based on the output result of the target ground subsidence prediction model so as to obtain the threshold value of the key factors influencing the ground subsidence, and the rapid early warning of the ground subsidence can be realized based on the threshold value of the real-time observation data corresponding to the key factors.
How to generate early warning information of ground subsidence in a targeted manner will be exemplarily described by means of fig. 7.
As shown in FIG. 7, the generation of the pre-warning information of the ground subsidence may include operations S710-S740.
In operation S710, feature observation data is constructed from the historical observation data of the plurality of dimensions and the real-time observation data of the plurality of dimensions based on the monte carlo algorithm.
The monte carlo algorithm is a method for solving the problem of statistical distribution and its critical value using random numbers, and its basic idea is to approximate the solution of a real problem by simulating a large number of random samples or processes, the general steps of which are as follows:
constructing a random probabilistic process such that the solution to the problem can be represented by certain parameters or characteristics of the process;
generating a plurality of random numbers from a known probability distribution as input or simulation of the process;
based on the results of the simulation, an estimate is calculated as an approximate solution to the problem.
In operation S720, the feature observation data is input to the target ground subsidence prediction model, and a prediction result corresponding to the feature observation data is obtained.
According to an embodiment of the present invention, the prediction result may include a ground deformation amount, a ground deformation rate, a deformation direction, and the like.
In operation S730, the feature observation data and the prediction result are obtained to characterize a target observation data type affecting the prediction result and an early warning threshold value corresponding to the target observation data type.
According to the embodiment of the invention, the neural network can be input according to the characteristics of the ground deformation and the early warning critical value for prediction so as to analyze and obtain the target observation data type and the corresponding early warning critical value for representing the influence prediction result.
According to the embodiment of the invention, the type of the observed data input into the target ground subsidence prediction model can be changed for a plurality of times, the main influencing factors are judged according to the prediction result output by the neural network, the factors such as the deformation rate, the groundwater level information, the deformation accumulation amount and the like can be kept unchanged, the value of the groundwater exploitation amount is changed and gradually changed to be large to judge whether the groundwater exploitation amount is the main influencing factor, if the prediction result output by the neural network is not changed greatly in the data value change process, the result is not the main influencing factor, otherwise, the result is judged to be the main influencing factor, and the early warning critical value corresponding to the target observed data type is divided according to the prediction result of the neural network. For example, when the underground water exploitation amount reaches 0.5 hundred million cubic meters, the underground water exploitation amount reaches a low-risk early warning critical value; when reaching 0.6 hundred million cubic meters, the warning threshold value of stroke danger is reached; reaching 0.7 hundred million cubic meters, the high risk early warning critical value is reached.
According to the embodiment of the invention, the early warning critical value can be divided into a plurality of stage critical values according to factors such as the damage degree and the influence range of ground deformation. For example, the low risk early warning threshold, the medium risk early warning threshold, the high risk early warning threshold, and the like can be classified. The foregoing examples are illustrative only and the invention is not limited to the division of the threshold values.
In operation S740, in response to the value of the real-time observation data corresponding to the target observation data type being greater than the pre-warning threshold, pre-warning information for characterizing that the target area has ground subsidence at a future time period is generated.
According to the embodiment of the invention, the early warning information of the response can be output according to the prediction result. For example, may be a warning level, including low risk warning, medium risk warning, and high risk warning. And the system can also comprise early warning information such as early warning areas, early warning time and the like. The foregoing pre-warning area may be a specific geographic location where ground subsidence occurs, for example, may be a geographic coordinate, or may be a range. The foregoing early warning time may include a time and duration of a possible occurrence of ground subsidence, and the like.
According to the embodiment of the invention, based on the Monte Carlo algorithm, the input of the model can be adjusted in a targeted manner according to the output result of the model, the analysis of the type of the sensitivity data of the observed data affecting the ground subsidence can be realized, and the threshold value of the more accurate sensitivity data can be obtained, so that the early warning critical value can be set in a targeted manner to predict the ground subsidence more accurately.
According to an exemplary embodiment of the present invention, a region is selected as the investigation region having an area of approximately 85.97 square kilometers. Firstly, 8 leveling point measurement data in a research area, inSAR data with 5m space precision and 2 geological drilling data in the research area are collected; and acquiring data such as hydrogeological parameters (permeability coefficient and water storage rate) of the I-IV water-bearing group in the research area and the change value of the annual groundwater level in 2020.
Secondly, calibrating InSAR data by using leveling data, and obtaining a 500m multiplied by 500m ground deformation grid data set by using a region average value method; acquiring drilling sand-viscosity and corresponding proportion values thereof based on a statistical method; and (3) obtaining 500m multiplied by 500m grid data of the research area by using a Kriging interpolation method through the water level difference, the permeability coefficient, the water storage rate and the sand-viscosity ratio. Unifying all the data into a file format, carrying out normalization processing by using the formula (1) to obtain normalized data, and selecting a learning set of the neural network model according to the correlation between the data and ground subsidence. Proper processing methods and parameters can be selected according to actual conditions so as to ensure the accuracy and usability of data.
Thirdly, clustering and dividing into a sedimentation center area, a sedimentation edge area, a sedimentation stable area and the like according to the distance and aggregation between the data points in the feature space; and then taking the ground subsidence value as a target value, dividing the characteristic data into a training set of 80% and a testing set of 20%, and training a ground parade subsidence prediction model by adopting a deep neural network structure formed by combining a convolutional neural network and a cyclic neural network. The neural network comprises 2 input neurons, 10 hidden layer neurons and 1 output neuron, the training maximum round is 5000, the learning rate is 0.1, and the training target error is 1×10 -5. After the model is pre-trained, a set Kalman filtering algorithm is adopted to process and fuse data, and the structure and parameters of the deep neural network model are corrected.
Finally, assuming that the ground subsidence prediction model is obtained by training the current water level, hydrogeologic parameters and other data as observation data, 1000 Monte Carlo and deep neural networks can be operated to perform simulation experiments, the ground subsidence is predicted, and the early warning critical value is divided. For example, a ground settlement low risk threshold (20 mm) and a high risk threshold (> 30 mm) may be set. The ground settlement risk area grades are defined according to the threshold value, and the ground settlement risk area grades comprise a low risk area (< 20 mm), a medium risk area (20 mm-30 mm), a high risk area (> 30 mm) and the like. The early warning time can be adjusted according to different divided early warning stages, potential influences of ground subsidence on infrastructure and environment are evaluated, and corresponding early warning information is output, such as building cracks, road collapse, pipeline breakage, groundwater level change and the like.
Based on the ground subsidence prediction method, the invention also provides a ground subsidence prediction device. The device will be described in detail below in connection with fig. 8.
Fig. 8 shows a block diagram of a ground subsidence prediction apparatus according to an embodiment of the present invention.
As shown in fig. 8, the ground subsidence prediction apparatus 800 of this embodiment includes a synchronization module 810, an adjustment module 820, and a prediction module 830.
The synchronization module 810 is configured to perform synchronization processing on the observation data in response to the received observation data for multiple dimensions of the target area in the current period, so as to obtain real-time observation data. In an embodiment, the synchronization module 810 may be configured to perform the operation S210 described above, which is not described herein.
The adjustment module 820 is configured to adjust model parameters of a pre-training model based on real-time observation data by using a kalman filtering algorithm to obtain a target ground subsidence prediction model, where the pre-training model is obtained by training an initial model by using historical observation data of multiple dimensions of a target area in a historical period. In an embodiment, the adjustment module 820 may be used to perform the operation S220 described above, which is not described herein.
The prediction module 830 is configured to input real-time observation data into a target ground subsidence prediction model, and output prediction information, where the prediction information characterizes whether ground subsidence occurs in a target area in a future period. In an embodiment, the prediction module 830 may be configured to perform the operation S230 described above, which is not described herein.
According to an embodiment of the invention, the adjustment module comprises: the device comprises a label acquisition sub-module, a covariance calculation sub-module, a Kalman gain matrix calculation sub-module and a parameter adjustment sub-module. The label obtaining sub-module is used for obtaining the ground subsidence predicted value output by the pre-training model in the training stage and the label corresponding to the ground subsidence predicted value; the covariance calculation sub-module is used for obtaining a covariance matrix of the ground subsidence prediction error according to the ground subsidence prediction value and the label corresponding to the ground subsidence prediction value; the Kalman gain matrix calculation sub-module is used for obtaining a Kalman gain matrix according to a covariance matrix of ground subsidence prediction errors, an observation data matrix and a covariance matrix of observation noise, wherein the observation data matrix is constructed according to real-time observation data and historical observation data; the covariance matrix of the observation noise is constructed according to the noise corresponding to the real-time observation data and the noise corresponding to the historical observation data; and the parameter adjustment sub-module is used for adjusting model parameters based on real-time observation data by utilizing the Kalman gain matrix to obtain a target ground subsidence prediction model.
According to an embodiment of the present invention, the real-time observation data includes S batches, S is an integer greater than 1, and the adjustment module further includes: a first covariance calculation sub-module and a first Kalman gain matrix calculation sub-module. The first covariance calculation sub-module is used for obtaining the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s+1st batch according to the Kalman gain matrix corresponding to the real-time observation data of the S-1st batch, the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the S-1st batch and the observation data matrix corresponding to the real-time observation data of the S-1st batch, wherein S is an integer greater than 1 and less than or equal to S-1. The first Kalman gain matrix calculation sub-module is used for obtaining a Kalman gain matrix corresponding to the real-time observation data of the s+1st batch according to the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s+1st batch, the observation data matrix corresponding to the real-time observation data of the s+1st batch and the covariance matrix of the observation noise corresponding to the real-time observation data of the s+1st batch. According to an embodiment of the invention, the synchronization module comprises: a temporal synchronization sub-module and a spatial synchronization sub-module. The time synchronization sub-module is used for performing time synchronization processing on the time-varying observation data in the observation data according to a preset time length scale to obtain time-synchronized real-time observation data; and the space synchronization sub-module is used for performing space synchronization processing on the observation data which changes with space in the observation data according to a preset space scale to obtain space synchronization real-time observation data.
According to the embodiment of the invention, the target ground subsidence prediction model comprises a convolution feature extraction module, a circulation feature extraction module and a prediction module; the prediction module comprises a convolution extraction sub-module, a circulation extraction sub-module and an information prediction sub-module. The convolution extraction submodule is used for extracting the spatial characteristics of the real-time observation data by utilizing the convolution characteristic extraction module; the cyclic extraction sub-module is used for processing the space characteristics by utilizing the cyclic characteristic extraction module to obtain time sequence characteristics of real-time observation data; and the prediction sub-module is used for processing the time sequence characteristics by utilizing the prediction module to obtain prediction information.
According to an embodiment of the present invention, the ground subsidence prediction apparatus further includes: the system comprises a feature observation data construction module, a result prediction module, a prediction result analysis module and an early warning information generation module. The characteristic observation data construction module is used for constructing characteristic observation data according to historical observation data of multiple dimensions and real-time observation data of multiple dimensions based on a Monte Carlo algorithm; the result prediction module is used for inputting the characteristic observation data into the target ground subsidence prediction model to obtain a prediction result corresponding to the characteristic observation data; the prediction result analysis module is used for analyzing the characteristic observation data and the prediction result to obtain a target observation data type for representing the influence prediction result and an early warning critical value corresponding to the target observation data type; and the early warning information generation module is used for generating early warning information used for representing ground subsidence of the target area in a future period in response to the fact that the value of the real-time observation data corresponding to the target observation data type is larger than an early warning critical value.
Any of the synchronization module 810, the adjustment module 820, and the prediction module 830 may be combined in one module to be implemented, or any of the modules may be split into a plurality of modules, according to an embodiment of the present invention. Or at least some of the functionality of one or more of the modules may be combined with, and implemented in, at least some of the functionality of other modules. At least one of the synchronization module 810, the adjustment module 820, and the prediction module 830 may be implemented at least in part as hardware circuitry, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system-on-chip, a system-on-a-substrate, a system-on-package, an Application Specific Integrated Circuit (ASIC), or in hardware or firmware, such as any other reasonable manner of integrating or packaging the circuitry, or in any one of or a suitable combination of three of software, hardware, and firmware, in accordance with embodiments of the present invention. Or at least one of the synchronization module 810, the adjustment module 820 and the prediction module 830 may be at least partially implemented as a computer program module which, when executed, performs the corresponding functions.
Fig. 9 shows a block diagram of a model training apparatus according to an embodiment of the present invention.
As shown in fig. 9, the ground subsidence model training subsidence prediction apparatus 900 of this embodiment includes an acquisition module 910, a data synchronization module 920, a processing module 930, a loss module 940, and a pre-training module 950.
The acquisition module 910 is configured to acquire sample historical observation data for a plurality of dimensions of a sample region over a historical period. In an embodiment, the obtaining module 910 may be configured to perform the operation S310 described above, which is not described herein.
The data synchronization module 920 is configured to perform synchronization processing on the sample history observation data to obtain target sample history observation data. In an embodiment, the data synchronization module 920 may be configured to perform the operation S320 described above, which is not described herein.
The processing module 930 is configured to input the historical observation data of the target sample into an initial model to obtain sample prediction information, where the sample prediction information characterizes a probability of occurrence of ground subsidence in the sample area in the target period. In an embodiment, the processing module 930 may be configured to perform the operation S330 described above, which is not described herein.
The loss module 940 is configured to obtain a loss value based on the loss function according to the sample prediction information and the sample label. In an embodiment, the loss module 940 may be used to perform the operation S340 described above, which is not described herein.
The pre-training module 950 is configured to adjust model parameters of the initial model based on the loss value to obtain a pre-trained model. In an embodiment, the pre-training module 950 may be configured to perform the operation S350 described above, which is not described herein.
Fig. 10 shows a block diagram of an electronic device adapted to implement a ground settlement prediction method according to an embodiment of the invention.
As shown in fig. 10, the electronic apparatus 1000 according to the embodiment of the present invention includes a processor 1001 that can perform various appropriate actions and processes according to a program stored in a ROM (Read Only Memory) or a program loaded from the storage section 1008 into a RAM (Random Access Memory ). The processor 1001 may include, for example, a general purpose microprocessor (e.g., a CPU), an instruction set processor and/or an associated chipset and/or a special purpose microprocessor (e.g., an Application Specific Integrated Circuit (ASIC)), or the like. The processor 1001 may also include on-board memory for caching purposes. The processor 1001 may include a single processing unit or a plurality of processing units for performing different actions of the method flow according to an embodiment of the invention.
In the RAM 1003, various programs and data necessary for the operation of the electronic apparatus 1000 are stored. The processor 1001, the ROM 1002, and the RAM 1003 are connected to each other by a bus 1004. The processor 1001 performs various operations of the method flow according to the embodiment of the present invention by executing programs in the ROM 1002 and/or the RAM 1003. Note that the program may be stored in one or more memories other than the ROM 1002 and the RAM 1003. The processor 1001 may also perform various operations of the method flow according to an embodiment of the present invention by executing programs stored in the one or more memories.
According to an embodiment of the invention, the electronic device 1000 may further comprise an I/O interface 1005, the I/O interface 1005 also being connected to the bus 1004. The electronic device 1000 may also include one or more of the following components connected to the I/O interface 1005: an input section 1006 including a keyboard, a mouse, and the like; an output portion 1007 including a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), etc., and a speaker, etc.; a storage portion 1008 including a hard disk or the like; and a communication section 1009 including a network interface card such as a LAN card, a modem, or the like. The communication section 1009 performs communication processing via a network such as the internet. The drive 1010 is also connected to the I/O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like, is installed as needed in the drive 1010, so that a computer program read out therefrom is installed as needed in the storage section 1008.
The present invention also provides a computer-readable storage medium that may be embodied in the apparatus/device/system described in the above embodiments; or may exist alone without being assembled into the apparatus/device/system. The computer-readable storage medium carries one or more programs which, when executed, implement methods in accordance with embodiments of the present invention.
According to embodiments of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, which may include, for example, but is not limited to: a portable computer diskette, a hard disk, a RAM (Random Access Memory ), a ROM (Read Only Memory), an erasable programmable Read-Only Memory (EPROM or flash Memory), a portable compact disc Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. For example, according to embodiments of the invention, the computer-readable storage medium may include ROM 1002 and/or RAM 1003 described above and/or one or more memories other than ROM 1002 and RAM 1003.
Embodiments of the present invention also include a computer program product comprising a computer program containing program code for performing the method shown in the flowcharts. The program code means for causing a computer system to carry out the method for predicting ground subsidence provided by the embodiments of the present invention when the computer program product is run on the computer system.
The above-described functions defined in the system/apparatus of the embodiment of the present invention are performed when the computer program is executed by the processor 1001. The systems, apparatus, modules, units, etc. described above may be implemented by computer program modules according to embodiments of the invention.
In one embodiment, the computer program may be based on a tangible storage medium such as an optical storage device, a magnetic storage device, or the like. In another embodiment, the computer program may also be transmitted in the form of signals on a network medium, distributed, and downloaded and installed via the communication section 1009, and/or installed from the removable medium 1011. The computer program may include program code that may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.
In such an embodiment, the computer program may be downloaded and installed from a network via the communication portion 1009, and/or installed from the removable medium 1011. The above-described functions defined in the system of the embodiment of the present invention are performed when the computer program is executed by the processor 1001. The systems, devices, apparatus, modules, units, etc. described above may be implemented by computer program modules according to embodiments of the invention.
According to embodiments of the present invention, program code for carrying out computer programs provided by embodiments of the present invention may be written in any combination of one or more programming languages, and in particular, such computer programs may be implemented in high-level procedural and/or object-oriented programming languages, and/or in assembly/machine languages. Programming languages include, but are not limited to, such as Java, c++, python, "C" or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).
The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
Those skilled in the art will appreciate that the features recited in the various embodiments of the invention and/or in the claims may be combined in various combinations and/or combinations, even if such combinations or combinations are not explicitly recited in the invention. In particular, the features recited in the various embodiments of the invention and/or in the claims can be combined in various combinations and/or combinations without departing from the spirit and teachings of the invention. All such combinations and/or combinations fall within the scope of the invention.
The embodiments of the present invention are described above. These examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Although the embodiments are described above separately, this does not mean that the measures in the embodiments cannot be used advantageously in combination. The scope of the invention is defined by the appended claims and equivalents thereof. Various alternatives and modifications can be made by those skilled in the art without departing from the scope of the invention, and such alternatives and modifications are intended to fall within the scope of the invention.
Claims (8)
1. A ground subsidence prediction method, comprising:
In response to received observation data aiming at a plurality of dimensions of a target area in a current period, synchronously processing the observation data to obtain real-time observation data;
Adjusting model parameters of a pre-training model based on the real-time observation data by utilizing a Kalman filtering algorithm to obtain a target ground subsidence prediction model, wherein the pre-training model is obtained by training an initial model by utilizing historical observation data of a plurality of dimensions of a target area in a historical period; and
Inputting the real-time observation data into the target ground subsidence prediction model, and outputting prediction information, wherein the prediction information characterizes the probability of ground subsidence of the target area in a future period;
The method comprises the steps of utilizing a Kalman filtering algorithm to adjust model parameters of a pre-training model based on real-time observation data to obtain a target ground subsidence prediction model, wherein the real-time observation data comprises S batches, S is an integer larger than 1, and the method comprises the following steps:
acquiring a ground subsidence predicted value output by the pre-training model in a training stage and a label corresponding to the ground subsidence predicted value;
obtaining a covariance matrix of ground subsidence prediction errors according to the ground subsidence prediction values and the labels corresponding to the ground subsidence prediction values;
Obtaining a Kalman gain matrix according to the covariance matrix of the ground settlement prediction error, the covariance matrix of the observation data matrix and the covariance matrix of the observation noise, wherein the observation data matrix is constructed according to the real-time observation data and the historical observation data; the covariance matrix of the observation noise is constructed according to the noise corresponding to the real-time observation data and the noise corresponding to the historical observation data;
adjusting the model parameters based on the real-time observation data by utilizing the Kalman gain matrix to obtain a target ground subsidence prediction model;
Obtaining a covariance matrix of a ground settlement prediction error corresponding to the real-time observation data of the s+1st batch according to a Kalman gain matrix corresponding to the real-time observation data of the S-1 st batch, a covariance matrix of a ground settlement prediction error corresponding to the real-time observation data of the S-1 st batch and an observation data matrix corresponding to the real-time observation data of the S-1 st batch, wherein S is an integer greater than 1 and less than or equal to S-1; and
And obtaining a Kalman gain matrix corresponding to the real-time observation data of the s+1st batch according to the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s+1st batch, the observation data matrix corresponding to the real-time observation data of the s+1st batch and the covariance matrix of the observation noise corresponding to the real-time observation data of the s+1st batch.
2. The method for predicting ground subsidence according to claim 1, wherein the synchronizing the observed data to obtain real-time observed data comprises:
according to a preset time length scale, performing time synchronization processing on the time-varying observation data in the observation data to obtain time-synchronized real-time observation data; and
And carrying out space synchronization processing on the observation data which changes with space in the observation data according to a preset space scale to obtain space synchronization real-time observation data.
3. The ground subsidence prediction method of claim 1, wherein the target ground subsidence prediction model comprises a convolution feature extraction module, a circulation feature extraction module, and a prediction module; inputting the real-time observation data into the target ground subsidence prediction model, and outputting prediction information, wherein the method comprises the following steps:
extracting spatial features of the real-time observation data by using the convolution feature extraction module;
Processing the spatial features by using the cyclic feature extraction module to obtain time sequence features of the real-time observation data; and
And processing the time sequence characteristics by using the prediction module to obtain the prediction information.
4. The method of ground subsidence prediction according to claim 1, further comprising:
based on a Monte Carlo algorithm, constructing characteristic observation data according to the historical observation data of the multiple dimensions and the real-time observation data of the multiple dimensions;
Inputting the characteristic observation data into the target ground subsidence prediction model to obtain a prediction result corresponding to the characteristic observation data;
The characteristic observation data and the prediction result are used for obtaining a target observation data type for representing influence on the prediction result and an early warning critical value corresponding to the target observation data type;
and generating early warning information used for representing that the ground subsidence occurs in the target area in a future period in response to the value of the real-time observed data corresponding to the target observed data type is larger than the early warning critical value.
5. A method of model training, comprising:
Acquiring sample historical observation data of a plurality of dimensions of a sample area in a historical period;
Synchronizing the sample historical observation data to obtain target sample historical observation data;
Inputting the historical observation data of the target sample into an initial model to obtain sample prediction information, wherein the sample prediction information characterizes the probability of ground subsidence of the sample area in a target period;
Obtaining a loss value according to the sample prediction information and the sample label based on the loss function; and
Model parameters of the initial model are adjusted based on the loss values to obtain a pre-trained model, wherein the pre-trained model is applied to the method of any one of claims 1-4.
6. A ground subsidence prediction apparatus, comprising:
The synchronization module is used for responding to the received observation data of multiple dimensions of the target area in the current period, and performing synchronization processing on the observation data to obtain real-time observation data;
the adjustment module is used for adjusting model parameters of a pre-training model based on real-time observation data by utilizing a Kalman filtering algorithm to obtain a target ground subsidence prediction model, wherein the pre-training model is obtained by training an initial model by utilizing historical observation data of a plurality of dimensions of the target area in a historical period; and
The prediction module is used for inputting the real-time observation data into the target ground subsidence prediction model and outputting prediction information, wherein the prediction information represents whether ground subsidence occurs in the target area in a future period;
wherein the real-time observation data includes S batches, S being an integer greater than 1, and the adjustment module includes:
The label obtaining sub-module is used for obtaining the ground subsidence predicted value output by the pre-training model in the training stage and the label corresponding to the ground subsidence predicted value;
The covariance calculation sub-module is used for obtaining a covariance matrix of the ground subsidence prediction error according to the ground subsidence prediction value and the label corresponding to the ground subsidence prediction value;
The Kalman gain matrix calculation sub-module is used for obtaining a Kalman gain matrix according to the covariance matrix of the ground subsidence prediction error, the covariance matrix of the observation data matrix and the observation noise, wherein the observation data matrix is constructed according to the real-time observation data and the historical observation data; the covariance matrix of the observation noise is constructed according to the noise corresponding to the real-time observation data and the noise corresponding to the historical observation data;
the parameter adjustment sub-module is used for adjusting the model parameters based on the real-time observation data by utilizing the Kalman gain matrix to obtain a target ground subsidence prediction model;
A first covariance calculation sub-module, configured to obtain a covariance matrix of a ground settlement prediction error corresponding to real-time observation data of an s+1st batch according to a kalman gain matrix corresponding to real-time observation data of the S-1st batch, a covariance matrix of the ground settlement prediction error corresponding to real-time observation data of the S-1st batch, and an observation data matrix corresponding to real-time observation data of the S-1st batch, where S is an integer greater than 1 and less than or equal to S-1;
The first Kalman gain matrix calculation sub-module is used for obtaining a Kalman gain matrix corresponding to the real-time observation data of the s+1st batch according to the covariance matrix of the ground settlement prediction error corresponding to the real-time observation data of the s+1st batch, the observation data matrix corresponding to the real-time observation data of the s+1st batch and the covariance matrix of the observation noise corresponding to the real-time observation data of the s+1st batch.
7. A model training device, comprising:
The acquisition module is used for acquiring sample historical observation data of a plurality of dimensions of the sample area in a historical period;
The data synchronization module is used for performing synchronization processing on the sample historical observation data to obtain target sample historical observation data;
The processing module is used for inputting the historical observation data of the target sample into an initial model to obtain sample prediction information, wherein the sample prediction information represents the probability of ground subsidence of the sample area in a target period;
The loss module is used for obtaining a loss value based on a loss function according to the sample prediction information and the sample label; and
A pre-training module for adjusting model parameters of the initial model based on the loss values to obtain a pre-trained model, wherein the pre-trained model is applied to the method of any one of claims 1-4.
8. An electronic device, comprising:
One or more processors;
Storage means for storing one or more programs,
Wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the method of any of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410121478.3A CN117648873B (en) | 2024-01-30 | 2024-01-30 | Ground subsidence prediction method, training method, device and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410121478.3A CN117648873B (en) | 2024-01-30 | 2024-01-30 | Ground subsidence prediction method, training method, device and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117648873A CN117648873A (en) | 2024-03-05 |
| CN117648873B true CN117648873B (en) | 2024-06-04 |
Family
ID=90048102
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410121478.3A Active CN117648873B (en) | 2024-01-30 | 2024-01-30 | Ground subsidence prediction method, training method, device and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117648873B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111563135A (en) * | 2020-03-31 | 2020-08-21 | 武汉大学 | GM (1,3) model urban ground subsidence prediction method combining terrain factor and neural network |
| CN112561191A (en) * | 2020-12-22 | 2021-03-26 | 北京百度网讯科技有限公司 | Prediction model training method, prediction method, device, apparatus, program, and medium |
| CN113095562A (en) * | 2021-04-07 | 2021-07-09 | 安徽天能清洁能源科技有限公司 | Ultra-short term power generation prediction method and device based on Kalman filtering and LSTM |
| CN113393032A (en) * | 2021-06-15 | 2021-09-14 | 西安电子科技大学 | Flight path cycle prediction method based on resampling |
| CN114881323A (en) * | 2022-05-07 | 2022-08-09 | 沈阳工业大学 | Foundation pit dewatering area underground water level prediction and updating method based on deep neural network |
| CN115077474A (en) * | 2022-06-14 | 2022-09-20 | 天津华北地质勘查总院 | Ground settlement trend prediction method and system based on machine learning |
| CN115526435A (en) * | 2022-11-09 | 2022-12-27 | 上海交通大学 | Oil well yield prediction method under spatio-temporal neural network model based on Kalman filtering |
| CN117074951A (en) * | 2023-07-18 | 2023-11-17 | 国网重庆市电力公司营销服务中心 | Lithium battery state of charge estimation method, device, equipment and readable storage medium |
-
2024
- 2024-01-30 CN CN202410121478.3A patent/CN117648873B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111563135A (en) * | 2020-03-31 | 2020-08-21 | 武汉大学 | GM (1,3) model urban ground subsidence prediction method combining terrain factor and neural network |
| CN112561191A (en) * | 2020-12-22 | 2021-03-26 | 北京百度网讯科技有限公司 | Prediction model training method, prediction method, device, apparatus, program, and medium |
| CN113095562A (en) * | 2021-04-07 | 2021-07-09 | 安徽天能清洁能源科技有限公司 | Ultra-short term power generation prediction method and device based on Kalman filtering and LSTM |
| CN113393032A (en) * | 2021-06-15 | 2021-09-14 | 西安电子科技大学 | Flight path cycle prediction method based on resampling |
| CN114881323A (en) * | 2022-05-07 | 2022-08-09 | 沈阳工业大学 | Foundation pit dewatering area underground water level prediction and updating method based on deep neural network |
| CN115077474A (en) * | 2022-06-14 | 2022-09-20 | 天津华北地质勘查总院 | Ground settlement trend prediction method and system based on machine learning |
| CN115526435A (en) * | 2022-11-09 | 2022-12-27 | 上海交通大学 | Oil well yield prediction method under spatio-temporal neural network model based on Kalman filtering |
| CN117074951A (en) * | 2023-07-18 | 2023-11-17 | 国网重庆市电力公司营销服务中心 | Lithium battery state of charge estimation method, device, equipment and readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117648873A (en) | 2024-03-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ghorbanzadeh et al. | A new GIS-based technique using an adaptive neuro-fuzzy inference system for land subsidence susceptibility mapping | |
| Wasowski et al. | Toward the next generation of research on earthquake-induced landslides: current issues and future challenges | |
| JP6635038B2 (en) | Simulation apparatus, simulation method, and storage medium | |
| McCreight et al. | Snow depth, density, and SWE estimates derived from GPS reflection data: Validation in the western US | |
| Zhao et al. | A Comparative Study of Landslide Susceptibility Mapping Using SVM and PSO‐SVM Models Based on Grid and Slope Units | |
| CN109085641B (en) | Monitoring data processing method for earthquake prediction, earthquake prediction method and system | |
| Boudevillain et al. | The Cévennes‐Vivarais Mediterranean Hydrometeorological Observatory database | |
| Corsini et al. | Use of ROC curves for early warning of landslide displacement rates in response to precipitation (Piagneto landslide, Northern Apennines, Italy) | |
| Frigerio et al. | A web-based platform for automatic and continuous landslide monitoring: The Rotolon (Eastern Italian Alps) case study | |
| Pudashine et al. | Deep learning for an improved prediction of rainfall retrievals from commercial microwave links | |
| Adnan Ikram et al. | Novel evolutionary-optimized neural network for predicting landslide susceptibility | |
| Pumo et al. | The SESAMO early warning system for rainfall-triggered landslides | |
| Dasgupta et al. | On the impacts of observation location, timing, and frequency on flood extent assimilation performance | |
| CN114493319A (en) | Cross-time-scale combined ancient landslide resurgence risk grading evaluation method and device | |
| Emberson et al. | Landslide hazard and exposure modelling in data‐poor regions: the example of the Rohingya refugee camps in Bangladesh | |
| Nickles et al. | How does the unique space‐time sampling of the SWOT mission influence river discharge series characteristics? | |
| Prakash et al. | A Bayesian approach to model selection and averaging of hydrostatic-season-temperature-time model | |
| Seo et al. | Objective reduction of rain gauge network via geostatistical analysis of uncertainty in radar-gauge precipitation estimation | |
| Vanden Boomen et al. | Application of deep learning for imaging‐based stream gaging | |
| CN117648873B (en) | Ground subsidence prediction method, training method, device and equipment | |
| Du et al. | Impact of assimilating airborne Doppler radar velocity data using the ARPS 3DVAR on the analysis and prediction of Hurricane Ike (2008) | |
| US20230130725A1 (en) | Systems and methods for encasement of advanced seismic sensors | |
| CN115795952A (en) | Slope deformation field reconstruction method based on finite elements and Bayesian kriging | |
| Pasari et al. | Nowcasting-Based Earthquake Hazard Estimation at Major Cities in New Zealand | |
| Kitterød | Estimating unconsolidated sediment cover thickness by using the horizontal distance to a bedrock outcrop as secondary information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20240322 Address after: No. 8, Haitai Huake 4th Road, Huayuan Industrial Zone, Binhai New Area, Tianjin 300450 Applicant after: ELECTRIC POWER SCIENCE & RESEARCH INSTITUTE OF STATE GRID TIANJIN ELECTRIC POWER Co. Country or region after: China Applicant after: STATE GRID TIANJIN ELECTRIC POWER Co. Applicant after: STATE GRID CORPORATION OF CHINA Address before: No. 8, Haitai Huake 4th Road, Huayuan Industrial Zone, Binhai New Area, Tianjin 300450 Applicant before: ELECTRIC POWER SCIENCE & RESEARCH INSTITUTE OF STATE GRID TIANJIN ELECTRIC POWER Co. Country or region before: China Applicant before: STATE GRID TIANJIN ELECTRIC POWER Co. Applicant before: STATE GRID CORPORATION OF CHINA Applicant before: Tianjin geological engineering survey and Design Institute Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |