CN113569479A - Long-term multi-step control method, device and storage medium for rock fracture development of stone cave temple - Google Patents
Long-term multi-step control method, device and storage medium for rock fracture development of stone cave temple Download PDFInfo
- Publication number
- CN113569479A CN113569479A CN202110848200.2A CN202110848200A CN113569479A CN 113569479 A CN113569479 A CN 113569479A CN 202110848200 A CN202110848200 A CN 202110848200A CN 113569479 A CN113569479 A CN 113569479A
- Authority
- CN
- China
- Prior art keywords
- time
- trend
- rock
- module
- time series
- 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.)
- Granted
Links
- 239000011435 rock Substances 0.000 title claims abstract description 87
- 206010017076 Fracture Diseases 0.000 title claims abstract description 52
- 238000000034 method Methods 0.000 title claims abstract description 50
- 239000004575 stone Substances 0.000 title claims abstract description 47
- 208000010392 Bone Fractures Diseases 0.000 title claims abstract description 41
- 238000011161 development Methods 0.000 title claims abstract description 33
- 230000007774 longterm Effects 0.000 title claims abstract description 30
- 238000003860 storage Methods 0.000 title claims description 15
- 230000000737 periodic effect Effects 0.000 claims abstract description 40
- 230000006870 function Effects 0.000 claims abstract description 16
- 230000004927 fusion Effects 0.000 claims abstract description 11
- 238000013135 deep learning Methods 0.000 claims abstract description 10
- 239000000284 extract Substances 0.000 claims abstract description 8
- 230000008859 change Effects 0.000 claims description 17
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 230000001687 destabilization Effects 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 230000008439 repair process Effects 0.000 claims description 4
- 238000007499 fusion processing Methods 0.000 claims description 3
- 230000008447 perception Effects 0.000 claims description 3
- 230000002787 reinforcement Effects 0.000 claims description 3
- 238000004891 communication Methods 0.000 claims 1
- 238000002474 experimental method Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 238000012544 monitoring process Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 5
- 230000000052 comparative effect Effects 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 239000000654 additive Substances 0.000 description 3
- 230000000996 additive effect Effects 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000010422 painting Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000001609 comparable effect Effects 0.000 description 1
- 238000012733 comparative method Methods 0.000 description 1
- 239000013078 crystal Substances 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000003340 mental effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000003449 preventive effect Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 230000001681 protective effect Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Business, Economics & Management (AREA)
- Mathematical Analysis (AREA)
- Quality & Reliability (AREA)
- Marketing (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Optimization (AREA)
- Entrepreneurship & Innovation (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Computational Linguistics (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Development Economics (AREA)
- Geometry (AREA)
- Game Theory and Decision Science (AREA)
- Probability & Statistics with Applications (AREA)
- Medical Informatics (AREA)
- Evolutionary Biology (AREA)
Abstract
The invention discloses a long-term multi-step control method and a device for rock fracture development of stone cave temple, wherein the method comprises the following steps: the MTLFnet network is provided based on the fusion of polynomial regression and deep learning, and the network describes the trend of the time sequence into a function with an independent variable as a time index for acquiring the multi-step time trend; the network consists of a data enhancement module, n periodic time sequence prediction modules and an uncertain time sequence prediction module; the periodic time sequence prediction module learns the decomposed single time sequence; and the uncertainty time sequence prediction module extracts the spatial dependence relationship among the multiple time sequences, and fuses the uncertainty time sequence time characteristics predicted by the polynomial regression and the spatial dependence relationship to obtain a prediction result of rock fracture development of the cave temple. The device comprises: a processor and a memory. The invention describes the trend of the time sequence into a function with an independent variable as a time index, is used for acquiring the multi-step time trend, and realizes long-term multi-step control on crack development.
Description
Technical Field
The invention relates to the field of stone cave temple rock mass cracks, in particular to a long-term multi-step control method, a long-term multi-step control device and a long-term multi-step control storage medium for stone cave temple rock mass crack development.
Background
The stone cave temple has wide distribution, large scale and complete system, integrates the arts of buildings, sculptures, wall paintings, calligraphy and the like, and fully embodies the aesthetic pursuit, value concept and cultural spirit of Chinese nationality. The stone cave temple has historical value, artistic value, cultural value, scientific value and social value at the same time, the stone cave temple depends on the engraving art for more than a thousand years, is the crystal of the hard work of the ancient artistic craftsman and is the precious mental wealth left by the predecessor. For the protection and utilization of the stone cave temple, China is still in the starting stage and still faces some outstanding problems: the safety risk of the stone cave temple cultural relics is high, and the stone cave temple cultural relics are easily influenced by geological disasters and biological diseases; the key protection technology for dealing with the diseases such as rock mass weathering, water seepage, structural instability and the like is not broken through. Therefore, the protection and utilization of the stone cave temple are enhanced, and the prediction of the rock fracture development of the stone cave temple becomes an urgent problem to be solved.
For the prediction task of rock fractures of stone cave temple, most of researches utilize a statistical method, a mechanical calculation method and a reliability analysis method, and in recent years, deep learning is also used for predicting the fractures of slope rock.
The traditional statistical models are: autoregressive (AR), Moving Average (MA) and Autoregressive Moving Average (ARMA), Autoregressive Integrated Moving Average (ARIMA), where Autoregressive Moving Average is a family of linear models that contain the above models. These statistical methods assume that there is a linear correlation between variables, and thus abstract the relationship between variables into a linear model.
In the field of time series data prediction at present, common machine learning models include: support Vector Regression (SVR), Deep Belief Networks (DBN), Random Forest Regression (RFR) and Recurrent Neural Networks (RNNs). The recurrent neural network RNN adds 'loops' in the network, so that the information can be more durable, the information is transmitted to the next judgment process after each judgment process of the neural network, the learning result at the previous moment is applied to the current learning process, and the problem of memory sustainability is solved. A Long-Short-Term-Memory network (LSTM) is derived on the basis of the RNN, and the problem that the effect of the RNN begins to be poor when the RNN depends on the Long distance is solved. Gated-round Unit (GRU) is another variation of RNN, which is structurally simpler than LSTM but has comparable effects to LSTM, both of which are widely used in time series prediction tasks.
The models have good effect in single variable time series prediction, but the models rarely focus on the prediction of a multivariable time series.
In recent years, multivariate timing prediction methods include self-adaptive shared memory Tensor LSTM (TLASM) and Long-and Short-term Time-series network (LSTNet). TLASM utilizes multitask 1dCNN to learn local time series characteristics, with the help of the self-adaptation shared memory, self-adaptation understanding the correlation between tasks, and according to this the degree of parameter sharing between the tasks is changed dynamically. LSTNet encodes short-term local information into low-dimensional vectors using a one-dimensional convolutional neural network, and decodes the vectors through a recurrent neural network. The method utilizes the advantages of the convolutional layer to find a dependency relationship mode between the local multi-dimensional input variable and the loop layer so as to capture complex long-term dependency relationship. Dual self-attention networks (dsanets) are used for efficient multivariate time series prediction without extraneous information. This illustrates that the use of the attention mechanism helps in time series prediction. The MTGNN model uses a graph neural network to capture spatial and temporal dependencies within a time series, applying the graph neural network to time series predictions. However, the above methods are only suitable for the case of large data set size, and when the data set size is small, the above methods have difficulty in learning the long-term trend of the sequence.
Over the millennium, the stone cave temple has been subject to various diseases both from nature and man-made. In order to strengthen digital protection on the cave temple and effectively solve the problem of structural instability caused by development of rock mass cracks of the cave temple, the inventor provides prediction on the development of the rock mass cracks of the cave temple. However, because the cave temple itself has the particularity of cultural relics, for the purpose of protection, systematic and large-scale data acquisition work has not been carried out in China, and standard specifications on aspects such as data acquisition, processing, storage, management and the like of the cave temple have not been established, so that monitoring data is difficult to share, and an open data set related to the aspects does not appear on a network. The absence of data means that the prediction effort starts with the collection of a data self-created dataset, which is a significant challenge for the inventors. Secondly, when the existing time series prediction method is used, a large amount of historical data is needed, monitoring of rock fractures of the stone cave temple is carried out late, a data set is small, changes of the rock fractures are slow, only changes of the fractures after a few hours or a few days are researched, the early warning significance of prediction of safety of the stone cave temple is not large, and changes of the rock fractures of the stone cave temple after a few months or a few years need to be predicted by an inventor. Meanwhile, rock mass fracture change is influenced by various meteorological factors, which means that the realization of rock mass fracture prediction faces the challenges of small data set, long-term prediction of multi-time-series prediction and long-term multi-step prediction of multi-time-series prediction.
Disclosure of Invention
The invention provides a long-term multistep control method, a device and a storage medium for fissure development of stone cave temple rock mass, which decompose a time sequence into a plurality of items based on data enhancement, indirectly enlarge data volume, and provide an MTLFnet network by fusing polynomial regression and deep learning, wherein the network describes the trend of the time sequence into a function with an independent variable as a time index, is used for obtaining multistep time trend, realizes long-term multistep control on fissure development, and is described in detail as follows:
in a first aspect, a long-term multi-step control method for rock fracture development of stone cave temple comprises the following steps:
the MTLFnet network is provided based on the fusion of polynomial regression and deep learning, and the network describes the trend of a time sequence into a function with an independent variable as a time index and is used for acquiring the multi-step time trend;
the network consists of a data enhancement module, n periodic time sequence prediction modules and an uncertainty time sequence prediction module;
the periodic time sequence prediction module learns the decomposed single time sequence; and the uncertainty time sequence prediction module extracts the spatial dependence relationship among the multiple time sequences, and fuses the uncertainty time sequence time characteristics predicted by the polynomial regression and the spatial dependence relationship to obtain a prediction result of rock fracture development of the cave temple.
Further, the periodic time series prediction module consists of a first polynomial regression module and two multilayer perception neural networks;
the uncertainty time sequence prediction module comprises a second polynomial regression module, a space convolution module, a splicing function and an MLP layer;
the periodic time series prediction module learns the decomposed single time series; and the space convolution module extracts the space dependence relationship among the multiple time sequences, and the MLP layer fuses the uncertainty time sequence time characteristics predicted by the polynomial regression and the space dependence relationship to obtain a prediction result.
In one embodiment, the uncertainty time series y is decomposed into trend terms y using L1 trend filteringtrAnd a random term yr;
Estimating the trend of all time points of the sequence yThe evaluation is combined into a trend sequence ytr,Will ytrAnd inputting the time characteristics of the uncertainty time series into a second polynomial regression module to predict the time characteristics of the uncertainty time series.
In one embodiment, the first polynomial regression module is used for capturing the time series variation with time, and describing the time index and the trend term of the time seriesThe future time index is input into a polynomial, and a multi-step trend of a single sequence is generated at one time
The first MLP and the second MLP respectively simulate the period of the time sequenceLaw of change of random termMLP of the last layer will trendPeriod of timeAnd random itemsFused into a time seriesN periodic time sequences are all subjected to fusion processing to obtainWill be provided withInputting into a spatial convolution module to learn spatial variation
In one embodiment, the trend of the uncertainty time series is predicted to be the resultSpatial variation predicted from periodic time seriesThe prediction results of rock fracture development of stone cave temple are obtained by fusion of the rock fracturesThe degree of the rock mass at the position is pre-judged whether the rock mass is in a destabilization state or not, and if the rock mass is in the destabilization state, repair and emergency reinforcement can be carried out in advance.
In one embodiment, the method further comprises: constructing a rock fracture data set of the stone cave temple,
the data set includes: time, width of rock mass fracture, temperature in the grotto, humidity in the grotto and grotto frost/dew point.
In a second aspect, a long-term multistep control device is developed in cave temple rock mass crack, the device includes: a processor and a memory, the memory having stored therein program instructions, the processor calling the program instructions stored in the memory to cause the apparatus to perform the method steps of any of the first aspects.
In a third aspect, a computer-readable storage medium storing a computer program comprising program instructions which, when executed by a processor, cause the processor to carry out the method steps of any one of the first aspect.
The technical scheme provided by the invention has the beneficial effects that:
1. by obtaining the prediction result of the development of the rock fractures of the stable rock temple for a long time, the rock temple manager can judge whether the state of the rock temple is stable, and when the rock temple is possibly unstable, early warning is given in advance, and precautionary measures are given in advance;
2. the invention can prevent the instability condition from worsening, and the protective capability of the stone cave temple can be comprehensively improved by combining the preventive protection with the rescue protection.
Drawings
FIG. 1 is a flow chart of a long-term multi-step control method for rock fracture development of a stone cave temple;
FIG. 2 is a schematic diagram of a periodic time series decomposition;
FIG. 3 is a schematic diagram of a periodic time series prediction;
FIG. 4 is a RMSE visualization;
fig. 5 is a schematic structural diagram of a cave temple crack development control device based on regression fusion deep learning.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention are described in further detail below.
In order to solve the problems, the embodiment of the invention firstly develops the monitoring activity of rock cave temple rock mass crack data in the round holes in the north rock cave of Qingyang city in Gansu province and the Anyue county in Sichuan province in 10 months in 2020, establishes a north rock cave rock mass crack data set, and is also a first rock cave temple rock mass crack monitoring data set in China. The data set can be used for researching the influence of meteorological factors on the development of rock fractures of stone cave temple.
Aiming at the problem of small data set, the embodiment of the invention provides a data enhancement method which decomposes a time sequence into a plurality of items and indirectly enlarges the data volume. The time sequence data which is greatly influenced by meteorological factors and time factors can show obvious periodicity, the trend development mode, the periodic mode and the random mode of the time sequence data are decomposed, modeling prediction is carried out on different modes according to different methods, and the development rule of the time sequence data can be better met.
Aiming at long-term multi-step prediction of multivariate time series data, the embodiment of the invention provides that a polynomial regression and a deep learning method are fused to provide an MTLFnet network, the network describes the trend of a time series into a function with an independent variable as a time index, and the multi-step time trend can be obtained once only by inputting the time index. And predicting the rock fracture change value within twenty days after three months on the rock fracture data set of the stone cave temple constructed in the embodiment of the invention by using MTLFnet. Compared with the existing time sequence prediction method, the comprehensive experiment result shows that the MFTLTnet obtains a better effect on a real rock mass fracture data set of the stone cave temple.
In the following, the embodiment of the invention will describe the long-term multi-step prediction network structure (MTLFnet) for rock fracture prediction of a stone cave temple in more detail. The MTLFnet model proposed by the embodiment of the present invention is shown in fig. 1. The MTLF network consists of a data enhancement module, n periodic time series prediction modules (PTFs), and an uncertainty time series prediction module. The periodic time series prediction module (PTF) is composed of three parts of polynomial regression and two multilayer perceptive neural networks (MLPs). The uncertainty time series prediction module includes a Polynomial regression module (Polynomial regression), a spatial convolution module (CONV), Cat () and an MLP layer in fig. 1. The periodic time sequence prediction module learns the decomposed single time sequence; and the space convolution module extracts the space dependence relationship among the multiple time sequences, and the MLP layer fuses the uncertainty time sequence time characteristics predicted by the polynomial regression and the space dependence relationship to obtain a prediction result.
Each time sequence in the multiple time sequences presents a change trend of different forms, and the time sequence data are divided into two categories for learning, wherein the two categories are a periodic time sequence and an uncertain time sequence respectively. The periodic time series means that the time series exhibits regular changes in a period of days, months, years, and the like. The uncertainty time series is understood as being a time series which cannot be determined whether a sequence has a period when the sequence length is short, and is called as the uncertainty time series.
MTLFnet separately learns each sequence, a single sequence adopts polynomial regression to simulate the trend change curve of the sequence, and the periodic rule and random change of the sequence are learned through MLP. And fusing the single time sequences by using 1dCNN, extracting the mutual relation among the sequences as the spatial characteristic of the target sequence, and mixing the long-term trend characteristic of the target sequence learned by polynomial regression to finally obtain the prediction result of the target sequence.
Data enhancement module
To achieve data enhancement, the periodic time series uses the search _ decompose function in the statmodels module to decompose each time node in the time series into trend (trend term), search (seasonal term), and residual term (or random term). The decomplexe data decomposition model mainly has two types: additive models (additive) and multiplicative models (multiplicative), and the additive models are used in the model.
Uncertainty time series y the series is decomposed into trend terms y using L1 trend filteringtrAnd a random term yr, y=ytr+yr. The objective function of the L1 trend filter is shown in equation (1):
wherein, λ can control the trend precision after decomposition, and the smaller λ is, the closer the trend is to the original sequence.Representing the trend estimate, y, of the time series at time ttRepresenting the original value at time t of the time series.
The trend estimated values of the sequence y at all time are combined into a sequence, and the sequence is the trend sequence ytr,Will ytrThe time characteristics of the uncertainty time series can be predicted by inputting to a Polynomial regression module (Polynomial regression module) in fig. 1.
Second, periodic time series prediction module
The periodic time series prediction module is also the PTF module in fig. 2, and there are n PTF modules in n periodic time series. The module consists of three parts, namely first polynomial regression and two multilayer perceptive neural networks (MLPs). This module is used to predict a single periodic time series, and the module is constructed as shown in fig. 3.
The first polynomial regression module in FIG. 3 captures the time series variation with time, and plots the time index and the trend term of the time seriesFunctional relationship between them. Inputting the future time index into the polynomial, a multi-step trend of a single sequence can be generated at a timeAnd no errors accumulate. The left side of fig. 3 has two MLP layers from top to bottom, which capture the non-linear variation of the time series. The first MLP and the second MLP respectively simulate the period of the time sequenceLaw of change of random termThe MLP of the last layer of FIG. 3 will trendPeriod of timeAnd random itemsFused into a time series That is, the prediction result of the ith periodic time series, this process can be expressed by equation (2):
all the n periodic time sequences are processed by equation (2), and the result is obtainedWill be provided withThe spatial variation can be learned by inputting the spatial convolution module CONV in the uncertainty time series prediction module.
Uncertainty time series prediction module
The uncertainty time sequence prediction module consists of two parts, namely a time characteristic description module and a time space fusion module of the uncertainty time sequence.
The time characteristic description module of the uncertain time series is arranged to learn the trend characteristic of the uncertain time series, and the module still relies on Polynomial regression, that is, the Polynomial regression module in fig. 1 realizes the process, which can be expressed as equation (3):
the input of the equation is a trend sequence y obtained in an s data enhancement moduletrThe result obtained by the equationIs a trend prediction result for an uncertainty time series.
The space-time fusion module comprises: a spatial convolution module (CONV), a splicing function Cat () and an MLP layer. Predicting the equation in the module(2) Inputting the obtained result into a space convolution module to obtain the predicted space change of the periodic time sequenceThis process can be expressed as equation (4):
the space-time fusion module predicts the trend of the uncertain time seriesSpatial variation predicted from periodic time seriesFusing to obtain the predicted resultThe process can be expressed as equation (5):
the result obtained in equation (5)Namely the prediction result of the fissure development of stone cave temple rock mass according toThe size of the rock mass can be used for prejudging the danger level of the rock mass at the position, whether the rock mass is in a destabilization state or not, and if the rock mass is in the destabilization state, repairing and emergency repair reinforcing can be carried out in advance.
Fourth, the experimental setup
In the experiments of the present example, the model was trained using Adam optimizer with learning rate set to 0.0001, the polynomial regression maximum order term to 12 times, the accuracy of the L1 trend filtering set to 0.2, the batch size 32, and the data window width 12 × 7, 12 × 10, 12 × 15. The time interval was set at 12 x 30 x 3 and represents data after 3 months of model prediction. The model comprises 3 periodic time sequence prediction modules and an uncertain sequence prediction module. Each cycle time series prediction module comprises one 1dCNN and two MLPs, and the 1dCNN has two convolution layers with maximum pooling.
Fifthly, data set and evaluation index
In 10 months in 2020, monitoring activities of rock cave temple rock mass fracture data are developed in the round-sense cave in Qingyang city of Gansu province and the Anyue county of Sichuan province, a north rock cave rock mass fracture data set is established, and the first rock cave temple rock mass fracture monitoring data set in China is also provided. The data set recorded the change data of one fissure in the north grove in 2020, 10/month, 22 to 2021, 4/month, 8 days, and monitored the temperature, humidity, frost/dew point in the grove at the same time. The data set can be used for researching the influence of meteorological factors on the development of rock fractures of stone cave temple.
Self-building a data set: five lines of data are concentrated in the rock fracture data of the stone cave temple, namely time, rock fracture width, temperature in the stone cave, humidity in the stone cave and frost/dew point in the stone cave. The sampling frequency of the fracture width is low, and the fracture width is collected every two hours. Meanwhile, meteorological factors influencing the change of the crack width are changed frequently, and the sampling frequency is once every 20 minutes. In order to predict the change of the fracture width, the data of the fracture width and the image data in the grotto need to be aligned, and then denoising processing is carried out. The data set is divided into a training set, a verification set and a test set according to the ratio of 6:2: 2.
In the experiment, two commonly used evaluation indexes, root Mean square error rmse (root Mean Squared error), Mean Absolute error mae (Mean Absolute error), were used. Both are defined as:
wherein, yi,Respectively representing a true value and a predicted value, and M represents the number of data points. The lower the values of the two indexes, RMSE and MAE, the better the prediction.
Sixth, comparison of Experimental Properties
In this section, the model proposed by the embodiments of the present invention is evaluated and compared with the current mainstream time series prediction algorithm, including methods based on deep learning, such as CNN, LSTM, CNNLSTM, and LSTnet. The experiment is carried out on the Chinese painting data set established by the embodiment of the invention. Table 1 shows the results of comparative experiments of 5 methods for predicting rock fractures of stone cave temple within 20 days after 2 months in the future. It can be seen that the MTLFnet model of the present invention is significantly superior to all previous methods, predicting a loss RMSE of 0.3076, lower than all comparative methods.
TABLE 15 comparative experimental results of the methods
Names | RMSE | MAE |
CNN | 1.5687 | 1.2663 |
LSTM | 2.0942 | 1.6712 |
CNNLSTM | 4.7126 | 0.9057 |
LSTnet | 2.0734 | 1.9677 |
MTLFnet | 0.3334 | 0.3076 |
In order to embody the superiority of long-term multi-step prediction of MTLFnet, the method performs prediction experiments on rock fractures of stone cave temple within 20 days in the future after 3 months, and the table 2 shows the comparative experiment results of 5 methods. Observing the results of each experiment in the table 2 after evaluation by two evaluation indexes of RMSE and MAE, it can be seen that the MTLFnet provided by the invention has stable performance and unchanged prediction performance after the prediction interval is prolonged, and is still superior to other methods.
TABLE 25 comparative experimental results of the methods
Names | RMSE | MAE |
CNN | 1.8999 | 1.4859 |
LSTM | 2.4985 | 1.2140 |
CNNLSTM | 14.0778 | 3.1149 |
LSTnet | 0.5236 | 0.4787 |
MTLFnet | 0.3399 | 0.3039 |
The results of RMSE visualization of the five models are shown in fig. 5, and are output every 5 epochs after 60 epochs, with the abscissa being the output times and the ordinate being the corresponding RMSE values. The dotted line in fig. 5 is the MFLTnet proposed by the present invention, and it can be seen that the MTLFnet model not only converges fast, but also has the least error compared to other methods.
In summary, the embodiment of the present invention provides a network architecture MTLFnet with polynomial regression and deep learning, which is used to learn the correlation between multiple time sequences, and implement long-term multi-step prediction on a single target sequence by using multiple time sequences. The core of the MTLFnet network lies in classifying time sequences, decomposing periodic time sequences and then respectively learning; and learning the sequence trend after filtering the uncertainty time sequence trend. The long-term change characteristics of the time sequence are simulated through polynomial regression, 1DCNN learns the hidden relation among the multiple time sequences, namely the spatial change characteristics, and the multi-step prediction results are generated once after the long-term characteristics and the spatial characteristics are input into an MLP layer.
Based on the same inventive concept, the embodiment of the invention also provides a rock fracture development control device of the stone cave temple based on the parallel hybrid network, referring to fig. 5, the device comprises: a processor 1 and a memory 2, the memory 2 having stored therein program instructions, the processor 1 calling the program instructions stored in the memory 2 to cause the apparatus to perform the following method steps in an embodiment:
the MTLFnet network is provided based on the fusion of polynomial regression and deep learning, and the network describes the trend of a time sequence into a function with an independent variable as a time index and is used for acquiring the multi-step time trend;
the network consists of a data enhancement module, n periodic time sequence prediction modules and an uncertain time sequence prediction module;
the periodic time sequence prediction module learns the decomposed single time sequence; and the uncertainty time sequence prediction module extracts the spatial dependence relationship among the multiple time sequences, and fuses the uncertainty time sequence time characteristics predicted by the polynomial regression and the spatial dependence relationship to obtain a prediction result of rock fracture development of the cave temple.
The periodic time series prediction module consists of a first polynomial regression module and two multilayer perception neural networks;
the uncertainty time sequence prediction module comprises a second polynomial regression module, a space convolution module, a splicing function and an MLP layer;
the periodic time sequence prediction module learns the decomposed single time sequence; and the space convolution module extracts the space dependence relationship among the multiple time sequences, and the MLP layer fuses the uncertainty time sequence time characteristics predicted by the polynomial regression and the space dependence relationship to obtain a prediction result.
Further, uncertainty time series y is decomposed into trend terms y using L1 trend filteringtrAnd a random term yr;
Merging the trend estimated values of all the time points of the sequence y into a trend sequence ytr,Will ytrAnd inputting the time characteristics of the uncertainty time series into a second polynomial regression module to predict the time characteristics of the uncertainty time series.
The first polynomial regression module is used for capturing the change of the time series along with the time and describing the trend items of the time index and the time seriesThe future time index is input into a polynomial, and a multi-step trend of a single sequence is generated at one time
The first MLP and the second MLP respectively simulate the period of the time sequenceLaw of change of random termMLP of the last layer will trendPeriod of timeAnd random itemsFused into a time seriesN periodic time sequences are all subjected to fusion processing to obtainWill be provided withInputting into a spatial convolution module to learn spatial variation
Go toStep by step, predicting the trend of the uncertainty time seriesSpatial variation predicted from periodic time seriesThe prediction results of rock fracture development of stone cave temple are obtained by fusion of the rock fracturesThe degree of the rock mass at the position is pre-judged whether the rock mass is in a destabilization state or not, and if the rock mass is in the destabilization state, repair and emergency reinforcement can be carried out in advance.
In one embodiment, the method further comprises: constructing a rock fracture data set of the stone cave temple,
the data set includes: time, width of rock mass fracture, temperature in the grotto, humidity in the grotto and grotto frost/dew point.
It should be noted that the device description in the above embodiments corresponds to the method description in the embodiments, and the embodiments of the present invention are not described herein again.
The execution main bodies of the processor 1 and the memory 2 may be devices having a calculation function, such as a computer, a single chip, a microcontroller, and the like, and in the specific implementation, the execution main bodies are not limited in the embodiment of the present invention, and are selected according to requirements in practical applications.
The memory 2 and the processor 1 transmit data signals through the bus 3, which is not described in detail in the embodiment of the present invention.
Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, where the storage medium includes a stored program, and when the program runs, the apparatus on which the storage medium is located is controlled to execute the method steps in the foregoing embodiments.
The computer readable storage medium includes, but is not limited to, flash memory, hard disk, solid state disk, and the like.
It should be noted that the descriptions of the readable storage medium in the above embodiments correspond to the descriptions of the method in the embodiments, and the descriptions of the embodiments of the present invention are not repeated here.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. The procedures or functions according to the embodiments of the invention are brought about in whole or in part when the computer program instructions are loaded and executed on a computer.
The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored on or transmitted over a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium or a semiconductor medium, etc.
In the embodiment of the present invention, except for the specific description of the model of each device, the model of other devices is not limited, as long as the device can perform the above functions.
Those skilled in the art will appreciate that the drawings are only schematic illustrations of preferred embodiments, and the above-described embodiments of the present invention are merely provided for description and do not represent the merits of the embodiments.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (8)
1. A long-term multi-step control method for rock fracture development of stone cave temple is characterized by comprising the following steps:
the MTLFnet network is provided based on the fusion of polynomial regression and deep learning, and the network describes the trend of a time sequence into a function with an independent variable as a time index and is used for acquiring the multi-step time trend;
the network consists of a data enhancement module, n periodic time sequence prediction modules and an uncertainty time sequence prediction module;
the periodic time sequence prediction module learns the decomposed single time sequence; and the uncertainty time sequence prediction module extracts the spatial dependence relationship among the multiple time sequences, and fuses the uncertainty time sequence time characteristics predicted by the polynomial regression and the spatial dependence relationship to obtain a prediction result of rock fracture development of the cave temple.
2. The long-term multi-step control method for rock fracture development of stone cave temple according to claim 1,
the periodic time series prediction module consists of a first polynomial regression module and two multilayer perception neural networks;
the uncertainty time sequence prediction module comprises a second polynomial regression module, a space convolution module, a splicing function and an MLP layer;
the periodic time series prediction module learns the decomposed single time series; and the space convolution module extracts the space dependence relationship among the multiple time sequences, and the MLP layer fuses the uncertainty time sequence time characteristics predicted by the polynomial regression and the space dependence relationship to obtain a prediction result.
3. The long-term multi-step control method for rock fracture development of stone cave temple according to claim 2,
the uncertainty time series y is decomposed into a trend term y using L1 trend filteringtrAnd a random term yr;
4. The long-term multi-step control method for rock fracture development of stone cave temple according to claim 2,
the first polynomial regression module is used for capturing the change of the time series along with the time and describing the trend items of the time index and the time seriesThe future time index is input into a polynomial, and a multi-step trend of a single sequence is generated at one time
The first MLP and the second MLP respectively simulate the period of the time sequenceLaw of change of random termMLP of the last layer will trendPeriod of timeAnd random itemsFused into a time seriesN periodic time sequences are all subjected to fusion processing to obtainWill be provided withInputting into a spatial convolution module to learn spatial variation
5. The long-term multi-step control method for rock fracture development of stone cave temple according to claim 2,
predicting the trend of the uncertainty time seriesSpatial variation predicted from periodic time seriesThe prediction results of rock fracture development of stone cave temple are obtained by fusion of the rock fracturesThe degree of the rock mass at the position is pre-judged whether the rock mass is in a destabilization state or not, and if the rock mass is in the destabilization state, repair and emergency reinforcement can be carried out in advance.
6. The long-term multi-step control method for rock fracture development of stone cave temple according to claim 5, further comprising: constructing a rock fracture data set of the stone cave temple,
the data set includes: time, width of rock mass fracture, temperature in the grotto, humidity in the grotto and grotto frost/dew point.
7. The utility model provides a long-term multistep controlling means is grown in rock cave temple rock mass crack, a serial communication port, the device includes: a processor and a memory, the memory having stored therein program instructions, the processor calling upon the program instructions stored in the memory to cause the apparatus to perform the method steps of any of claims 1-6.
8. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program comprising program instructions which, when executed by a processor, cause the processor to carry out the method steps of any of claims 1-6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110848200.2A CN113569479B (en) | 2021-07-27 | 2021-07-27 | Long-term multi-step control method, device and storage medium for rock mass crack development of stone cave temple |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110848200.2A CN113569479B (en) | 2021-07-27 | 2021-07-27 | Long-term multi-step control method, device and storage medium for rock mass crack development of stone cave temple |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113569479A true CN113569479A (en) | 2021-10-29 |
CN113569479B CN113569479B (en) | 2023-11-10 |
Family
ID=78167710
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110848200.2A Active CN113569479B (en) | 2021-07-27 | 2021-07-27 | Long-term multi-step control method, device and storage medium for rock mass crack development of stone cave temple |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113569479B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115808944A (en) * | 2023-02-09 | 2023-03-17 | 国能大渡河枕头坝发电有限公司 | Stator temperature rise test control method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101677008B1 (en) * | 2015-11-20 | 2016-11-17 | (주)엔키아 | Method for Data Forecasting based on Time Series Decomposition |
CN110610232A (en) * | 2019-09-11 | 2019-12-24 | 南通大学 | Long-term and short-term traffic flow prediction model construction method based on deep learning |
CN112101660A (en) * | 2020-09-15 | 2020-12-18 | 重庆交通大学 | Rainfall type landslide displacement prediction model and method based on staged attention mechanism |
US20210064689A1 (en) * | 2019-08-27 | 2021-03-04 | Nec Laboratories America, Inc. | Unsupervised multivariate time series trend detection for group behavior analysis |
CN112949896A (en) * | 2020-12-22 | 2021-06-11 | 浙江佰安信息技术有限公司 | Time sequence prediction method based on fusion sequence decomposition and space-time convolution |
CN113065703A (en) * | 2021-03-31 | 2021-07-02 | 天津大学 | Time series prediction method combining multiple models |
-
2021
- 2021-07-27 CN CN202110848200.2A patent/CN113569479B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101677008B1 (en) * | 2015-11-20 | 2016-11-17 | (주)엔키아 | Method for Data Forecasting based on Time Series Decomposition |
US20210064689A1 (en) * | 2019-08-27 | 2021-03-04 | Nec Laboratories America, Inc. | Unsupervised multivariate time series trend detection for group behavior analysis |
CN110610232A (en) * | 2019-09-11 | 2019-12-24 | 南通大学 | Long-term and short-term traffic flow prediction model construction method based on deep learning |
CN112101660A (en) * | 2020-09-15 | 2020-12-18 | 重庆交通大学 | Rainfall type landslide displacement prediction model and method based on staged attention mechanism |
CN112949896A (en) * | 2020-12-22 | 2021-06-11 | 浙江佰安信息技术有限公司 | Time sequence prediction method based on fusion sequence decomposition and space-time convolution |
CN113065703A (en) * | 2021-03-31 | 2021-07-02 | 天津大学 | Time series prediction method combining multiple models |
Non-Patent Citations (1)
Title |
---|
张振;李黎;田莹;周嘉陵;李媛;顾嘉伟;: "连续运行参考站网的暴雨过程时空变化分析", 测绘科学, no. 08 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115808944A (en) * | 2023-02-09 | 2023-03-17 | 国能大渡河枕头坝发电有限公司 | Stator temperature rise test control method |
CN115808944B (en) * | 2023-02-09 | 2023-06-02 | 国能大渡河枕头坝发电有限公司 | Stator temperature rise test control method |
Also Published As
Publication number | Publication date |
---|---|
CN113569479B (en) | 2023-11-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2019022854A1 (en) | Data2data: deep learning for time series representation and retrieval | |
CN110674604A (en) | Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM | |
WO2018012487A1 (en) | Forecasting device, parameter set production method and program | |
CN102945240A (en) | Method and device for realizing association rule mining algorithm supporting distributed computation | |
CN107622303A (en) | For the method for neutral net and the equipment of execution this method | |
JP6775469B2 (en) | OD traffic predictors, methods, and programs | |
US11954115B2 (en) | Methods and systems for wavelet based representation | |
CN110909125B (en) | Detection method of media rumor of news-level society | |
CN112163671A (en) | New energy scene generation method and system | |
CN113411216A (en) | Network flow prediction method based on discrete wavelet transform and FA-ELM | |
Mobin et al. | Multi-objective X-bar control chart design by integrating NSGA-II and data envelopment analysis | |
CN114694379B (en) | Traffic flow prediction method and system based on self-adaptive dynamic graph convolution | |
CN113569479A (en) | Long-term multi-step control method, device and storage medium for rock fracture development of stone cave temple | |
WO2022098601A1 (en) | Autonomous fluid management using fluid digital twins | |
US20220164688A1 (en) | System and method for automated imputation for multi-state sensor data and outliers | |
JP6398991B2 (en) | Model estimation apparatus, method and program | |
Bogaerts et al. | A fast inverse approach for the quantification of set-theoretical uncertainty | |
CN112348175A (en) | Method for performing feature engineering based on reinforcement learning | |
CN113570135B (en) | Parallel hybrid network-based grotto temple rock mass crack development control method and device | |
Minaei-Bidgoli et al. | Correlation mining between time series stream and event stream | |
Ghosh et al. | Statistical analysis of landscape data: Space-for-time, probability surfaces and discovering species | |
CN117851928A (en) | Method and system for predicting shale oil yield based on CNN-LSTM model | |
CN108614887B (en) | Internet of things information processing method and system with noise in supply chain environment | |
Vikranth et al. | PRICE PREDICTION SYSTEM-A PREDICTIVE DATA ANALYTICS USING ARIMA MODEL. | |
Saini et al. | A Comparison Between Long Short-Term Memory And Prophet For Time Series Analysis And Forecasting Technique |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |