CN117744504A - Flood discharge atomization rain intensity analysis model building method and device - Google Patents
Flood discharge atomization rain intensity analysis model building method and device Download PDFInfo
- Publication number
- CN117744504A CN117744504A CN202410186436.8A CN202410186436A CN117744504A CN 117744504 A CN117744504 A CN 117744504A CN 202410186436 A CN202410186436 A CN 202410186436A CN 117744504 A CN117744504 A CN 117744504A
- Authority
- CN
- China
- Prior art keywords
- flood discharge
- data
- analysis model
- rain intensity
- atomization
- 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.)
- Pending
Links
- 238000004458 analytical method Methods 0.000 title claims abstract description 104
- 238000000889 atomisation Methods 0.000 title claims abstract description 73
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000012549 training Methods 0.000 claims abstract description 79
- 238000013528 artificial neural network Methods 0.000 claims abstract description 33
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 30
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 16
- 238000011144 upstream manufacturing Methods 0.000 claims abstract description 8
- 230000007787 long-term memory Effects 0.000 claims abstract description 6
- 230000015654 memory Effects 0.000 claims description 35
- 239000003595 mist Substances 0.000 claims description 16
- 238000011176 pooling Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000003860 storage Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 7
- 238000009499 grossing Methods 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 abstract description 7
- 230000006403 short-term memory Effects 0.000 abstract description 5
- 230000008569 process Effects 0.000 description 9
- 238000013135 deep learning Methods 0.000 description 7
- 238000003062 neural network model Methods 0.000 description 6
- 230000009467 reduction Effects 0.000 description 6
- 230000009286 beneficial effect Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 5
- 238000009792 diffusion process Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000007726 management method Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 230000010485 coping Effects 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000007921 spray Substances 0.000 description 1
- 210000002105 tongue Anatomy 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000013316 zoning 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
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a flood discharge atomization rain intensity analysis model building method and device, and relates to the technical field of flood discharge monitoring. The method comprises the following steps: constructing an initial analysis model based on a convolutional neural network and a long-term and short-term memory neural network, and acquiring sample flood discharge data comprising upstream and downstream water level difference, flood discharge amount, hole pattern, trajectory and atomized rain intensity; and inputting the sample flood discharge data into the initial analysis model for iterative training to obtain a flood discharge atomization rain intensity analysis model. Aiming at the problem that the existing analysis means is difficult to summarize the complex relationship between different flood discharge conditions and atomized rain, the invention utilizes the capacity of the convolutional neural network for extracting spatial features and the capacity of the long-term memory neural network for modeling time sequences, and adapts a prediction model to the complexity between different flood discharge conditions and the atomized rain intensity through flood discharge sample data, thereby accurately predicting the flood discharge atomized rain intensity and providing effective decision support for flood discharge management.
Description
Technical Field
The invention relates to the technical field of flood discharge monitoring, in particular to a method and a device for establishing a flood discharge atomization rain intensity analysis model.
Background
The flood discharge atomization rain is rainfall with range, specificity and randomness generated by the collision of the water tongues during the flood discharge of the dam. In the atomization rainfall process, flood discharge conditions (such as physical conditions of flood discharge devices such as a dam) are the basis of analysis of the atomization rainfall intensity. At present, the analysis and prediction of the flood discharge atomization rain is mainly performed according to an empirical formula proposed by a researcher and the empirical relationship among the upstream and downstream water level difference, the flood discharge amount and the atomization rain intensity of the flood discharge device. The method has better reliability for the flood discharge of the arch dam, but is difficult to summarize through an empirical formula in the face of more complex relations between flood discharge conditions and the intensity of the flood discharge atomization rain, for example, the performance of a diversion flood discharge hydropower station which is developed vigorously in southwest areas in China is obviously reduced according to the past experience prediction.
It has also been proposed by the learner to quantify the intensity of the atomized rain using a BP neural network, which is an approximation of a linear weight function in nature, to use the variation of some linear parameters in the function to perform some approximation on the pattern or function to be identified, which can be effective when there is no associated data before and after identification, but when there is associated data before and after identification, the algorithm cannot mine the pattern generated by the time series data. Moreover, the BP neural network generally needs a large amount of training data, especially in the complex problem, if insufficient atomized rain intensity data are not trained, the model may be difficult to generalize to a new condition, so that performance is reduced, an optimal solution of the model cannot be obtained, and accurate prediction of the atomized rain intensity is realized.
Therefore, in the field of flood discharge atomization rain analysis, an analysis method with higher efficiency, strong adaptability and high prediction accuracy is needed.
Disclosure of Invention
The invention provides a method and a device for establishing a flood discharge atomization rain intensity analysis model, which are used for solving the problem that the existing analysis means are difficult to accurately predict the atomization rain intensity according to the complex relationship between flood discharge conditions and the flood discharge atomization rain intensity.
The invention is realized by the following technical scheme:
the invention provides a flood discharge atomization rain intensity analysis model building method, which comprises the following steps:
s1, constructing an initial analysis model: an initial analysis model is built based on a convolutional neural network and a long-short-term memory neural network, the convolutional neural network is used for extracting spatial features of flood discharge data, the long-short-term memory neural network is used for predicting the spatial features to obtain atomized rain intensity, and the atomized rain intensity comprises transverse length, longitudinal length and height of atomized rain;
s2, training the initial analysis model: obtaining sample flood discharge data, wherein the sample flood discharge data comprises upstream and downstream water level difference, flood discharge amount, hole pattern, trajectory picking angle and atomization rain intensity; and inputting the sample flood discharge data into the initial analysis model for iterative training to obtain a flood discharge atomization rain intensity analysis model.
Aiming at the problem that the complex relation between different flood discharge conditions and atomized rain is difficult to summarize by an empirical formula, the complex relation between different flood discharge conditions and atomized rain is learned by an analysis model based on fusion of a convolutional neural network and a long-short-period memory neural network, the convolutional neural network has excellent extraction capacity of space features, the long-short-period memory neural network models time sequences, and a prediction model is trained under the condition that atomized rain intensity data are limited, so that the prediction model can adapt to the complexity between different flood discharge conditions and the atomized rain intensity of flood discharge, and the space features and the time features of the flood discharge conditions are fused, thereby accurately predicting the atomized rain intensity of flood discharge.
In one embodiment, the convolutional neural network comprises 2 convolutional layers and 2 pooling layers, the convolutional layers having a convolutional kernel size of 20×20 and a number of channels of 1.
In one embodiment, the long-short-term memory neural network includes a first LSTM layer and a second LSTM layer, where the first LSTM layer and the second LSTM layer are used to learn the spatial features to obtain output features combining the spatial features and the time sequence features, and an output of the second LSTM layer is sequentially connected with a dropout layer, a smoothing layer and a fully-connected layer.
In one embodiment, inputting the sample flood discharge data into the initial analytical model for iterative training comprises:
s211, processing the sample flood discharge data into characteristic data with the dimension of 4 and the length of N, inputting the characteristic data into the initial analysis model, and obtaining model output; the 4 dimensions of the characteristic data correspond to water level difference, flood discharge amount, hole type data and a cantilever flow angle in the flood discharge data, and the data length corresponds to the number of samples;
s212, taking the atomized rain intensity data corresponding to each sample as a sample label, and outputting and adjusting model parameters according to the sample label and the model to complete one-time iterative training;
s213, repeating S211-S212 until the training times reach a preset value or the model accuracy meets a preset condition.
In one embodiment, the method further comprises: setting a training period, acquiring flood discharge data in the current training period, and if the number of the flood discharge data in the current training period is larger than a preset value, inputting the flood discharge data in the current training period and historical flood discharge data into the flood discharge atomization rain intensity analysis model for optimization training, wherein the historical flood discharge data is acquired before the current training period.
In one embodiment, the method further comprises: obtaining flood discharge data under specific flood discharge conditions, wherein the specific flood discharge conditions comprise extreme meteorological conditions; and inputting the flood discharge data under the specific flood discharge condition into the flood discharge atomization rain intensity analysis for fine adjustment training to obtain an atomization rain intensity analysis model suitable for the specific flood discharge condition.
In a second aspect of the present invention, there is provided a method of analyzing the intensity of flood discharge atomized rain, the method comprising: acquiring real-time flood discharge data, and predicting the real-time flood discharge data through an atomized rain intensity analysis model to obtain atomized rain intensity; the atomization rain intensity analysis model is obtained by the flood discharge atomization rain intensity analysis model establishment method according to any embodiment.
In a third aspect of the present invention, there is provided a flood discharge atomization rain intensity analysis model building apparatus, the apparatus comprising:
the initial model building module is used for building an initial analysis model: an initial analysis model is built based on a convolutional neural network and a long-short-term memory neural network, the convolutional neural network is used for extracting spatial features of flood discharge data, the long-short-term memory neural network is used for predicting the spatial features to obtain atomized rain intensity, and the atomized rain intensity comprises transverse length, longitudinal length and height of atomized rain;
a training module for training the initial analysis model: obtaining sample flood discharge data, wherein the sample flood discharge data comprises upstream and downstream water level difference, flood discharge amount, hole pattern, trajectory picking angle and atomization rain intensity; and inputting the sample flood discharge data into the initial analysis model for iterative training to obtain a flood discharge atomization rain intensity analysis model.
In a fourth aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor implements the method for building a model for analyzing intensity of flood discharge mist and rain or the method for analyzing intensity of flood discharge mist and rain according to any one of the above embodiments when executing the computer program.
In a fifth aspect of the present invention, there is provided a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method for establishing a model for analyzing intensity of flood discharge mist rain or the method for analyzing intensity of flood discharge mist rain according to any of the above embodiments.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the convolutional neural network has excellent capacity of extracting spatial features and capacity of modeling a time sequence, the prediction model is adapted to the complexity between different flood discharge conditions and the intensity of the flood discharge atomized rain through limited training data, and the spatial features and the temporal features of the flood discharge conditions are fused, so that the intensity of the flood discharge atomized rain is accurately predicted;
2. the size of the convolution kernel is 20 multiplied by 20, the window size of the feature extracted from the input data by the model is defined, the larger convolution kernel can capture a larger range of features, is suitable for the problem related to long-range dependence or large-scale structure, can capture some large-scale features in flood discharge data in the input data, is beneficial to better capturing the relation between the features, has the convolution and channel number of 1, and considers the complexity of the problem and the balance of calculation resources;
3. sample flood discharge data are processed into characteristic data with the dimension of 4 and the length of the sample number as model input, and a mode of iterative training is adopted by batch input, so that the characteristics of data characteristics and data quantity of the flood discharge data are considered, and the model training efficiency is improved under limited training resources;
4. the method can perform model training and optimization under specific flood discharge conditions according to different flood discharge devices, geographical environments and meteorological conditions, so that the method is better suitable for various situations, and the decision capability of a flood discharge manager in coping with the change of the flood discharge conditions is enhanced.
Drawings
In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings that are needed in the examples will be briefly described below, it being understood that the following drawings only illustrate some examples of the present invention and therefore should not be considered as limiting the scope, and that other related drawings may be obtained from these drawings without inventive effort for a person skilled in the art. In the drawings:
FIG. 1 is a flow chart of a method for establishing a flood discharge atomization rain intensity analysis model according to an embodiment of the invention;
FIG. 2 is a schematic diagram of a convolutional neural network structure according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a long-short term memory neural network according to an embodiment of the present invention;
FIG. 4 is a graph of loss of CNN-LSTM according to an embodiment of the invention;
FIG. 5 is a graph showing the comparison of predicted values and actual monitored values of the flood discharge atomization rain intensity analysis model in a storm area;
fig. 6 is a graph comparing predicted values with real monitored values of the flood discharge atomization rain intensity analysis model in a mist zone.
Detailed Description
For the purpose of making apparent the objects, technical solutions and advantages of the present invention, the present invention will be further described in detail with reference to the following examples and the accompanying drawings, wherein the exemplary embodiments of the present invention and the descriptions thereof are for illustrating the present invention only and are not to be construed as limiting the present invention.
It is noted that the terms "comprising" and "having," and any variations thereof, in the description and claims of the present invention and in the foregoing figures, are intended to cover a non-exclusive inclusion, such as a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to or includes other steps or elements inherent to the apparatus.
The terminology used in the various embodiments of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the various embodiments of the application. As used herein, the singular is intended to include the plural as well, unless the context clearly indicates otherwise. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which various embodiments of this application belong. The terms (such as those defined in commonly used dictionaries) will be interpreted as having a meaning that is identical to the meaning of the context in the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein in connection with the various embodiments.
The embodiment of the invention provides a flood discharge atomization rain intensity analysis model building method, which is suitable for analyzing atomization rain intensity under different flood discharge conditions and is beneficial to improving the prediction accuracy of the atomization rain intensity and the application range.
As shown in fig. 1, fig. 1 is a flowchart of a method for establishing a flood discharge atomization rain intensity analysis model according to an embodiment of the present invention, including the following steps:
s1, constructing an initial analysis model: constructing an initial analysis model based on a convolutional neural network and a long-term and short-term memory neural network, wherein the convolutional neural network is used for extracting spatial features of flood discharge data, and the long-term and short-term memory neural network is used for predicting the spatial features to obtain the atomized rain intensity which comprises an atomized rain range;
s2, training the initial analysis model: acquiring historical flood discharge data, wherein the historical flood discharge data comprises upstream and downstream water level difference, flood discharge amount, hole pattern, diversion picking angle and atomization rain intensity; and inputting the historical flood discharge data into an initial analysis model, and performing iterative training on the initial analysis model to obtain a flood discharge atomization rain intensity analysis model.
The spatial characteristics of the data are processed using convolutional neural networks (Convolutional Neural Network, CNN) which extract spatial information in the data through convolutional and pooling layers to better understand the rain intensity distribution in the surrounding area of the dam. As shown in fig. 2, which is a schematic diagram of the convolutional neural network structure, the CNN mainly includes a convolutional layer and a pooling layer, which are very effective in extracting data features and spatial information. The convolution layer uses convolution check input data to carry out convolution operation so as to extract space characteristics in the rain intensity data, thereby being capable of better understanding the rain intensity distribution of the area around the flood discharge opening of the dam. Through the convolved feature map, the CNN can capture key features of the input data, providing powerful support for subsequent analysis and prediction.
Therefore, in the method, the selection of the size and the channel number of the convolution kernel is important to ensure that the CNN can accurately extract and analyze key features of the rain intensity data around the dam, thereby improving the accuracy of flood discharge decisions.
As a preferred embodiment, parameters for setting the CNN network structure are as follows: the CNN includes 2 convolution layers and 2 pooling layers, the convolution layers and pooling layers are alternately connected, the convolution kernel of the convolution layers is 20×20, and the number of channels is 1. The pooling area of the pooling layer is 2×2, the channel is 1, and the setting of the pooling layer corresponds to the setting of the convolution layer.
By analyzing the existing atomization rain intensity data, the factors influencing the atomization rain intensity are obviously spatially related, and also show different change trends in different time periods. Long short-term memory neural networks (LSTM) are used to process time series data to take into account the time dependence of rain intensity. As shown in fig. 3, which is a schematic diagram of a long-short-term memory neural network structure, the LSTM network can capture the sequence relationship in the data and predict the future rain intensity, and the LSTM is a special cyclic neural network, which is different from the traditional cyclic neural network and is mainly characterized by maintenance and control of the cell state. In dam flood discharge atomized rain intensity analysis, the LSTM neural network functions to capture the time dependence of the rain intensity data for better understanding and predicting future rain intensity changes. The cellular state of LSTM resembles a conveyor belt with only a small amount of linear interactions, which enables it to maintain data memory for long periods of time without forgetting or confusion. In addition, LSTM uses various structures called "gates" to control signal propagation to adjust the content of the cell state, thereby enabling the addition or deletion of data in the cell state. This feature makes LSTM very effective in the processing of time series data, helping to improve the accuracy and timeliness of dam flood discharge decisions. The number of layers and the time step of the LSTM network are configured to adapt to the time dependence of the data.
As a preferred embodiment, the parameters for setting the LSTM network structure are as follows: the long-term and short-term memory neural network comprises a first LSTM layer and a second LSTM layer, wherein the first LSTM layer and the second LSTM layer are used for learning the spatial characteristics to obtain output characteristics combining the spatial characteristics and the time sequence characteristics, and the output of the second LSTM layer is sequentially connected with a dropoff layer, a smoothing layer and a full-connection layer, and the dropoff layer is used for preventing overfitting.
In this embodiment, the first LSTM layer receives data output by the convolutional neural network, and performs timing feature learning. The first LSTM layer is mainly responsible for learning time sequence characteristics in input data, the LSTM is provided with a memory unit, can capture and store long-term time sequence dependency, and through time steps, the LSTM network can effectively process time sequence information and transfer the learned time sequence characteristics to the next layer. The second LSTM layer receives timing characteristics outputs from the first LSTM layer, which outputs include the results of the learning of the timing characteristics by the first LSTM layer. The main task of the second LSTM layer is to learn the spatial features, which will further integrate the temporal features from the first LSTM layer while taking into account the influence of the spatial features. Thus, LSTM networks can effectively capture timing and spatial correlation in data. And finally, outputting information containing the combination of time sequence and space characteristics by the second LSTM layer, and providing more comprehensive characteristic representation for the processing of the subsequent layers.
S2 is a model training and optimizing step, which comprises training a CNN-LSTM model by using historical flood discharge data, ensuring the reliability of the model through continuous cross validation and performance evaluation, wherein the training and optimizing of the model can adapt to continuously changing flood discharge conditions, such as physical conditions, geographical environment conditions, meteorological conditions and the like of a flood discharge device, so as to provide reliable flood discharge decision support.
The following is a training implementation step of a CNN-LSTM-based flood discharge atomization rain intensity analysis model, which comprises the following steps:
s201, data acquisition and preparation: the method comprises the steps of collecting historical flood discharge data of different flood discharge conditions, namely flood discharge data of different flood discharge devices, different geographical environments and different meteorological conditions, classifying the data sources, setting corresponding labels, wherein the flood discharge data comprise water level differences, flood discharge amount, hole type data, picking angles and the like, but the method is not limited to the water level differences, the flood discharge amount, the hole type data, the picking angles and the like, and specific flood discharge data can be selected and added according to the different flood discharge devices, the geographical environments and the meteorological conditions, but the corresponding historical atomization rain intensity data are necessary in the data so as to correct model output.
The data acquisition also includes acquiring real-time data for subsequent continuous optimization of the model.
The data preparation also includes the step of preprocessing the data, including denoising, formatting, normalizing, etc., of the data to ensure the quality and consistency of the data.
S202, setting model training parameters: setting the number of training or training termination conditions, optimizing algorithms, gradient thresholds, initial learning rate, learning rate reduction period, learning rate reduction factor, regularization parameters, etc.
In one embodiment, in the using process of the model, the model is continuously optimized according to the real-time collected data, so that flood discharge data in the current training period is obtained through setting a training period, if the number of the flood discharge data in the current training period is larger than a preset value, the flood discharge data in the current training period and the historical flood discharge data are input into the flood discharge atomization rain intensity analysis model for optimization training, and the historical flood discharge data are obtained before the current training period.
Because atomized rain sample data is not easy to collect and limited in quantity, in order to ensure that the model prediction performance is reduced due to the change of the flood discharge conditions, the geographic environment, the climate and the like over time, new sample data is acquired in the use process of the model to optimize and train the model, so that the model keeps continuously high prediction performance. By judging the number of collected samples in the period, the problem that training resources are wasted because the number of newly added sample data in the period is too small and the model is subjected to ineffective optimization once even when the set period is reached is avoided. In addition, when the number of samples newly increased in the period is enough, the model may be optimized by using only the number of samples newly increased.
S203, performing model training, comprising the following steps:
s211, processing sample flood discharge data into characteristic data with the dimension of 4 and the length of N, inputting the characteristic data into the initial analysis model, and obtaining model output, wherein the 4 dimensions of the characteristic data correspond to water level difference, flood discharge amount, hole type data and a picking angle in the flood discharge data, and the data length corresponds to the number of samples, namely the data amount of the flood discharge data acquired for different flood discharge conditions;
s212, taking the atomized rain intensity data corresponding to each sample as a sample label, and outputting and adjusting model parameters according to the sample label and the model to complete one-time iterative training. In the step, sample data are input into an initial model in batches for training, one batch of input is used as one-time training, and model parameters are trained in multiple iterations until training times or model accuracy meets preset conditions;
s213, repeating S211-S212 until the training times reach a preset value or the model accuracy meets a preset condition;
in one embodiment, the historical flood discharge data is divided into training data and test data before training, the initial analysis model is trained by the training data, the trained model is subjected to accuracy verification by the test data, the verification result meets the requirements, an atomized rain intensity analysis model is obtained, and the flood discharge data collected in real time is predicted by the model to obtain predicted atomized rain intensity data.
In one embodiment, when the model is applied to the atomized rain analysis under the specific flood discharge condition, the atomized rain intensity analysis model trained in the step S2 is subjected to fine tuning training by using the flood discharge data related to the specific flood discharge condition, so that the model is adapted to continuously and stably predict the atomized rain intensity under the condition. The method comprises the following steps: and obtaining flood discharge data under specific flood discharge conditions, inputting the flood discharge data under the specific flood discharge conditions into the analysis of the intensity of the atomized rain for fine adjustment training, and obtaining an analysis model of the intensity of the atomized rain suitable for the specific flood discharge conditions, wherein the specific flood discharge conditions comprise extreme meteorological conditions, special geographic environments, specific flood discharge devices and the like.
In a second aspect of the present invention, there is provided a method for analyzing the intensity of flood discharge atomized rain, comprising: acquiring real-time flood discharge data, and predicting the real-time flood discharge data through an atomized rain intensity analysis model to obtain atomized rain intensity; the atomization rain intensity analysis model is obtained by the flood discharge atomization rain intensity analysis model building method according to any embodiment of the invention.
Further, the dam flood discharge decision is supported by using the atomized rain intensity prediction data of the flood discharge atomized rain intensity analysis model. Based on the predicted change of the rain intensity, the flood risk flood discharge atomization rain intensity is evaluated, and whether flood discharge is carried out or not is determined so as to reduce risks such as flood.
In order to verify the prediction performance of the flood discharge atomization rain intensity analysis model, flood discharge data are input into CNN-LSTM trained by the method to obtain an atomization rain intensity prediction result and analyze the result as follows.
1. The basic settings of the CNN-LSTM model are as follows:
the CNN-LSTM neural network model is built by means of programming software, the model adopts 2 convolution layers and 2 pooling layers, a characteristic learning LSTM Layer and an LSTM output Layer, a dropout-Layer prevents overfitting, and finally a smoothing and full-connection Layer. Wherein the convolution kernel is 20×20, the number of channels is 1, the pooling area of the pooling layer is 2×2, and the channels are 1.
The training sample data is obtained by selecting 4 pieces of monitoring real data with the length of 256 of a built hydropower station in China, carrying out 2-layer convolution and pooling processing to obtain 1 piece of input data characteristic data with the length of 64, wherein the data has characteristic values of 4 pieces of data before compression, and has all the characteristics of 4 pieces of real data with the length of 256.
2. Training parameters of the CNN-LSTM model were set as follows:
1) Selecting an Adam optimization algorithm;
2) Maximum training period Max epochs=800;
3) Gradient threshold Gradient Threshold =1;
4) Initial learning rate Initial Learn Rate =0.01;
5) Setting the initial learning rate to be variable Learn Rate Schedule =piece wise;
6) Learning rate reduction cycle Learn Rate Drop Period =40;
7) Learning rate reduction factor Learn Rate Drop Factor =0.8;
8) L2Regularization parameter sets L2Regularization = 1e-3.
In practical cases, the CNN-LSTM network automatically adjusts the learning rate according to the set initial learning rate by a learning rate reduction factor and a learning rate reduction period, so that the learning rate is kept changing at any time, so as to adapt to the change of the response sample of the system and the difficulty of the sample, and L2regularization is set to prevent the overfitting phenomenon so as to ensure the correctness of the predicted value.
3. The training process of the CNN-LSTM neural network model is as follows:
based on actual measurement data of large hydropower stations in China, training a CNN-LSTM deep learning neural network. According to the method, 4 pieces of measured monitoring data with the length of 256 are selected as input data of a model, 1 piece of data with the length of 256 are selected as output data, namely 256 groups of atomization range data generated under different working conditions of monitoring the atomization rain of hydropower stations in China are collected, wherein 4 influencing factors which mainly influence the diffusion of the atomization rain are water head difference, flood discharge amount, hole patterns and drift angle. Training is carried out by using the first 232 measured data CNN-LSTM deep learning neural network, and reliability verification is carried out by using the last 24 measured data.
The training period is 8 rounds, each round iterates 100 times, the total number of the iterations is 800 for 232 monitoring input samples, the mean square error curve of the sample training is known, the mean square error at the initial training stage of the CNN-LSTM deep learning neural network is rapidly reduced, the mean square error tends to be stable when the iteration number reaches about 100, and the fluctuation is kept about 0.1. FIG. 4 is a graph of CNN-LSTM loss, showing a rapid decrease in loss magnitude, with a loss of 0.01 after only 20 iterations, and then the loss remains unchanged as the number of iterations increases. The CNN-LSTM deep learning neural network learning rate is 0.002048. In general, the spray diffusion model based on the atomization rain actual measurement data has small error, so that the trained atomization rain diffusion model can be verified and analyzed on the basis.
4. The prediction results of the CNN-LSTM neural network are as follows:
based on the training result of the CNN-LSTM deep learning neural network model, the prediction analysis is carried out on the last 24 atomization rain monitoring measured data, and the region of the simulation rainfall is simply divided into a storm region and a mist region due to obvious zoning phenomenon of atomization rainfall diffusion, but the simulation model can be decomposed into single random data for a plurality of regions, and the deep learning neural network model can predict the single random data. Sample parameters for the predictive analysis of the atomized rain diffusion model are shown in table 1.
TABLE 1 CNN-LSTM neural network training model validation sample
And inputting known parameters such as a basic condition water head (m), a flood discharge amount (m < mu >/s), a hole shape coefficient, a picking angle (degree) and the like of a dam flood discharge to be predicted by utilizing the trained CNN-LSTM deep learning neural network model, and predicting the range and the intensity of atomized rain generated by the dam flood discharge during flood discharge.
Fig. 5 is a graph of the predicted value and the actual monitored value of CNN-LSTM in a storm area, and fig. 6 is a graph of the predicted value and the actual monitored value of CNN-LSTM in a mist area, as can be seen from the graph: for the storm atomizing area, the prediction curve and the real curve of the storm atomizing area show high coincidence, the error is smaller, the error value of the transverse width prediction value of the storm atomizing area at the No. 7 flood discharge hole is maximum, the absolute error is 34m, the relative error is 23.2%, the relative error of the rest prediction points is within 10%, and the prediction value of the storm atomizing area is accurate. For the mist atomization area, the matching degree of the prediction curve and the real curve is slightly reduced compared with that of the storm atomization area, the error of the longitudinal length prediction value is maximum at the position of the No. 18 flood discharge hole, the absolute error is 51m, the relative error is 11.3%, the rest longitudinal length prediction errors are within 10%, the error of the transverse length prediction value is maximum at the position of the No. 7 flood discharge hole, the relative error is 9.2%, and in general, the error of the mist atomization area is not more than 13%, and the mist prediction value is accurate.
Compared with the prior art, the invention has the following beneficial effects:
1. the flood discharge safety is improved: by means of more accurate analysis and prediction of the rain intensity, the risk in the flood discharge process is reduced, so that the safety of a flood discharge area is improved, and possible damage caused by flood is reduced;
2. more accurate flood discharge decisions: by comprehensively considering the time and space correlation, the technology can provide more accurate flood discharge decision support. The method is helpful for reducing the risk of false alarm flood or delaying the possibility of flood discharge, and improves the reliability of flood discharge decision;
3. better adaptability: the method can be trained and optimized according to different flood discharge devices, geographical environments and meteorological conditions, so that the method is better suitable for various situations. This enhances the decision-making ability of the dam manager in coping with meteorological changes and changes in dam characteristics;
4. automated decision support: the technology can monitor the rain intensity data in real time, automatically provide decision support according to the output of the model, reduce the subjective intervention of operators, improve the efficiency of decision making, and reduce the economic and environmental losses caused by flood events and the risks of personnel life safety;
5. the flood discharge management efficiency is improved: the technology is beneficial to automation and intellectualization of flood discharge management, and the decision making process is more timely and accurate. This improves the efficiency and reliability of management.
In summary, the invention provides a more efficient, more accurate and more reliable analysis method for the intensity of the flood discharge atomization rain by combining the capability of the convolution-long-short-time memory neural network, and the invention makes remarkable progress in flood discharge management and flood discharge decision.
In a third aspect of the present invention, there is provided a flood discharge atomization rain intensity analysis model building apparatus, comprising:
the initial model building module is used for building an initial analysis model: constructing an initial analysis model based on a convolutional neural network and a long-short-term memory neural network, wherein the convolutional neural network is used for extracting spatial features of flood discharge data, and the long-short-term memory neural network is used for predicting the spatial features to obtain the atomized rain intensity, and the atomized rain intensity comprises the transverse length, the longitudinal length and the height of atomized rain;
a training module for training the initial analysis model: obtaining sample flood discharge data, wherein the sample flood discharge data comprises upstream and downstream water level difference, flood discharge amount, hole pattern, diversion picking angle and atomization rain intensity; and inputting the sample flood discharge data into the initial analysis model for iterative training to obtain a flood discharge atomization rain intensity analysis model.
In a fourth aspect of the present invention, there is provided an electronic apparatus including a processor, a memory, an input device, an output device, and a communication device; the number of processors in the computer device may be one or more; the processor, memory, input devices, and output devices in the electronic device may be connected by a bus or other means.
The memory is a computer-readable storage medium that can be used to store software programs, computer-executable programs, and modules. The processor executes various functional applications and data processing of the electronic device by running software programs, instructions and modules stored in the memory, so as to implement the flood discharge atomization rain intensity analysis model building method or the flood discharge atomization rain intensity analysis method according to any of the embodiments of the present invention.
The memory may mainly include a memory program area and a memory data area, wherein the memory program area may store an operating system, at least one application program required for a function; the storage data area may store data created according to the use of the terminal, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some examples, the memory may further include memory remotely located with respect to the processor, the remote memory being connectable to the electronic device through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input device 42 may be used to receive training data or data to be predicted, set parameter data, control instructions, etc.; the output device 43 outputs the predicted value, the intermediate processing result, and the like.
In a fifth aspect of the present invention, there is provided a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the flood discharge mist rain intensity analysis model building method or the flood discharge mist rain intensity analysis method of any of the embodiments of the present invention. The storage medium may be ROM/RAM, magnetic disk, optical disk, etc.
The foregoing description of the embodiments has been provided for the purpose of illustrating the general principles of the invention, and is not meant to limit the scope of the invention, but to limit the invention to the particular embodiments, and any modifications, equivalents, improvements, etc. that fall within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. The method for establishing the flood discharge atomization rain intensity analysis model is characterized by comprising the following steps of:
s1, constructing an initial analysis model: an initial analysis model is built based on a convolutional neural network and a long-short-term memory neural network, the convolutional neural network is used for extracting spatial features of flood discharge data, the long-short-term memory neural network is used for predicting the spatial features to obtain atomized rain intensity, and the atomized rain intensity comprises transverse length, longitudinal length and height of atomized rain;
s2, training the initial analysis model: obtaining sample flood discharge data, wherein the sample flood discharge data comprises upstream and downstream water level difference, flood discharge amount, hole pattern, trajectory picking angle and atomization rain intensity; and inputting the sample flood discharge data into the initial analysis model for iterative training to obtain a flood discharge atomization rain intensity analysis model.
2. The method for establishing the flood discharge atomization rain intensity analysis model according to claim 1, wherein the convolutional neural network comprises 2 convolutional layers and 2 pooling layers, the convolutional layers have a convolutional kernel size of 20×20, and the number of channels is 1.
3. The method for establishing the flood discharge atomization rain intensity analysis model according to claim 1, wherein the long-term memory neural network comprises a first LSTM layer and a second LSTM layer, the first LSTM layer and the second LSTM layer are used for learning the spatial characteristics to obtain output characteristics combining the spatial characteristics and the time sequence characteristics, and the output of the second LSTM layer is sequentially connected with a dropout layer, a smoothing layer and a full-connection layer.
4. The method for establishing the flood discharge atomization rain intensity analysis model according to claim 1, wherein the step of inputting the sample flood discharge data into the initial analysis model for iterative training comprises the following steps:
s211, processing the sample flood discharge data into characteristic data with the dimension of 4 and the length of N, inputting the characteristic data into the initial analysis model, and obtaining model output; the 4 dimensions of the characteristic data correspond to water level difference, flood discharge amount, hole type data and a cantilever flow angle in the flood discharge data, and the data length corresponds to the number of samples;
s212, taking the atomized rain intensity data corresponding to each sample as a sample label, and outputting and adjusting model parameters according to the sample label and the model to complete one-time iterative training;
s213, repeating S211-S212 until the training times reach a preset value or the model accuracy meets a preset condition.
5. The flood discharge atomized rain intensity analysis model building method according to claim 1, further comprising: setting a training period, acquiring flood discharge data in the current training period, and if the number of the flood discharge data in the current training period is larger than a preset value, inputting the flood discharge data in the current training period and historical flood discharge data into the flood discharge atomization rain intensity analysis model for optimization training, wherein the historical flood discharge data is acquired before the current training period.
6. The flood discharge atomized rain intensity analysis model building method according to claim 1, further comprising: obtaining flood discharge data under specific flood discharge conditions, wherein the specific flood discharge conditions comprise extreme meteorological conditions; and inputting the flood discharge data under the specific flood discharge condition into the flood discharge atomization rain intensity analysis for fine adjustment training to obtain an atomization rain intensity analysis model suitable for the specific flood discharge condition.
7. A method for analyzing the intensity of flood discharge atomized rain, comprising: acquiring real-time flood discharge data, and predicting the real-time flood discharge data through an atomized rain intensity analysis model to obtain atomized rain intensity; the atomized rain intensity analysis model is obtained by the flood discharge atomized rain intensity analysis model building method according to any one of claims 1-6.
8. Flood discharge atomization rain intensity analysis model building device is characterized in that the device comprises:
the initial model building module is used for building an initial analysis model: an initial analysis model is built based on a convolutional neural network and a long-short-term memory neural network, the convolutional neural network is used for extracting spatial features of flood discharge data, the long-short-term memory neural network is used for predicting the spatial features to obtain atomized rain intensity, and the atomized rain intensity comprises transverse length, longitudinal length and height of atomized rain;
a training module for training the initial analysis model: obtaining sample flood discharge data, wherein the sample flood discharge data comprises upstream and downstream water level difference, flood discharge amount, hole pattern, trajectory picking angle and atomization rain intensity; and inputting the sample flood discharge data into the initial analysis model for iterative training to obtain a flood discharge atomization rain intensity analysis model.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the flood discharge mist rain intensity analysis model establishment method according to any one of claims 1 to 6 or the flood discharge mist rain intensity analysis method according to claim 7 when executing the computer program.
10. A computer-readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements the flood discharge mist rain intensity analysis model establishment method according to any one of claims 1 to 6 or the flood discharge mist rain intensity analysis method according to claim 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410186436.8A CN117744504A (en) | 2024-02-20 | 2024-02-20 | Flood discharge atomization rain intensity analysis model building method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410186436.8A CN117744504A (en) | 2024-02-20 | 2024-02-20 | Flood discharge atomization rain intensity analysis model building method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117744504A true CN117744504A (en) | 2024-03-22 |
Family
ID=90261193
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410186436.8A Pending CN117744504A (en) | 2024-02-20 | 2024-02-20 | Flood discharge atomization rain intensity analysis model building method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117744504A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647980A (en) * | 2019-09-18 | 2020-01-03 | 成都理工大学 | Time sequence prediction method based on GRU neural network |
| WO2021109644A1 (en) * | 2019-12-06 | 2021-06-10 | 北京理工大学 | Hybrid vehicle working condition prediction method based on meta-learning |
| CN115099497A (en) * | 2022-06-28 | 2022-09-23 | 中国水利水电科学研究院 | CNN-LSTM-based real-time flood forecasting intelligent method |
| KR20230052440A (en) * | 2021-10-13 | 2023-04-20 | 배재대학교 산학협력단 | Method and apparatus for implementing cnn-based water level prediction model |
| CN116989749A (en) * | 2023-03-13 | 2023-11-03 | 福建中锐汉鼎数字科技有限公司 | Drainage basin flood prediction method based on artificial intelligent hydrologic model |
| CN117520784A (en) * | 2023-11-30 | 2024-02-06 | 中国船舶集团国际工程有限公司 | Groundwater level multi-step prediction method based on convolution attention long-short-term neural network |
-
2024
- 2024-02-20 CN CN202410186436.8A patent/CN117744504A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647980A (en) * | 2019-09-18 | 2020-01-03 | 成都理工大学 | Time sequence prediction method based on GRU neural network |
| WO2021109644A1 (en) * | 2019-12-06 | 2021-06-10 | 北京理工大学 | Hybrid vehicle working condition prediction method based on meta-learning |
| KR20230052440A (en) * | 2021-10-13 | 2023-04-20 | 배재대학교 산학협력단 | Method and apparatus for implementing cnn-based water level prediction model |
| CN115099497A (en) * | 2022-06-28 | 2022-09-23 | 中国水利水电科学研究院 | CNN-LSTM-based real-time flood forecasting intelligent method |
| CN116989749A (en) * | 2023-03-13 | 2023-11-03 | 福建中锐汉鼎数字科技有限公司 | Drainage basin flood prediction method based on artificial intelligent hydrologic model |
| CN117520784A (en) * | 2023-11-30 | 2024-02-06 | 中国船舶集团国际工程有限公司 | Groundwater level multi-step prediction method based on convolution attention long-short-term neural network |
Non-Patent Citations (7)
| Title |
|---|
| 彭新民;林芝;黄财元;刘昉;: "挑流泄洪雾化的人工神经网络模型初探", 中国农村水利水电, no. 01, 25 January 2006 (2006-01-25), pages 60 - 61 * |
| 徐源浩;邬强;李常青;陈游倩;张力;冉广;胡彩虹;: "基于长短时记忆(LSTM)神经网络的黄河中游洪水过程模拟及预报", 北京师范大学学报(自然科学版), no. 03, 15 June 2020 (2020-06-15), pages 387 - 393 * |
| 戴丽荣 等: "挑流泄洪雾化影响范围的人工神经网络模型预测", 水利水电技术, vol. 34, no. 05, 20 May 2005 (2005-05-20), pages 7 - 9 * |
| 戴丽荣, 张云芳, 张华, 刘媛媛: "挑流泄洪雾化影响范围的人工神经网络模型预测", 水利水电技术, no. 05, 20 May 2003 (2003-05-20) * |
| 林云;: "BP神经网络预测挑流泄洪雾化范围的应用分析", 水利科技与经济, no. 06, 30 June 2009 (2009-06-30), pages 521 - 522 * |
| 柳海涛, 孙双科, 刘之平, 王晓松: "泄洪雾化预测的人工神经网络方法探讨", 水利学报, no. 10, 28 October 2005 (2005-10-28), pages 1241 - 1245 * |
| 沈位刚;时梦楠;徐圣捷;: "基于遗传神经网络的泄洪雾化雨预测研究", 中国水运(下半月), no. 10, 15 October 2015 (2015-10-15), pages 155 - 157 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106529814B (en) | Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain | |
| CN111343650B (en) | Urban scale wireless service flow prediction method based on cross-domain data and loss resistance | |
| CN113177652A (en) | Photovoltaic output prediction method | |
| de Guia et al. | Using stacked long short term memory with principal component analysis for short term prediction of solar irradiance based on weather patterns | |
| CN113240184B (en) | Building space unit cold load prediction method and system based on federal learning | |
| CN113449919B (en) | Power consumption prediction method and system based on feature and trend perception | |
| Liao et al. | Ultra-short-term interval prediction of wind power based on graph neural network and improved bootstrap technique | |
| CN109242150A (en) | A kind of electric network reliability prediction technique | |
| CN114357670A (en) | Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder | |
| CN113516304A (en) | Space-time joint prediction method and device for regional pollutants based on space-time graph network | |
| CN115169724A (en) | Runoff prediction method based on space-time graph convolutional neural network | |
| CN114492978A (en) | Time-space sequence prediction method and device based on multi-layer attention mechanism | |
| CN114169434A (en) | Load prediction method | |
| Bellagarda et al. | Effectiveness of neural networks and transfer learning to forecast photovoltaic power production | |
| CN113595798A (en) | Network flow prediction method and system for improving lightning connection process optimization algorithm | |
| CN116109007B (en) | Power generation power determination method, server and storage medium | |
| CN117744504A (en) | Flood discharge atomization rain intensity analysis model building method and device | |
| CN117033923A (en) | Method and system for predicting crime quantity based on interpretable machine learning | |
| CN115616333A (en) | Power distribution network line loss prediction method and system | |
| CN113723670B (en) | Photovoltaic power generation power short-term prediction method with variable time window | |
| CN114372640A (en) | Wind power prediction method based on fluctuation sequence classification correction | |
| CN115563848A (en) | Distributed photovoltaic total radiation prediction method and system based on deep learning | |
| CN112633565A (en) | Photovoltaic power aggregation interval prediction method | |
| CN112787882A (en) | Internet of things edge traffic prediction method, device and equipment | |
| Lestari et al. | Rainfall Prediction using Spatial Convolutional Neural Networks and Recurrent Neural Networks |
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 |