Intelligent analysis of seepage data of small reservoirs based on rainfall characteristics

Abstract

The application discloses small reservoir seepage data intelligent analysis based on rainfall characteristics, generates corresponding features through feature construction on seepage pressure data, rainfall and reservoir water level, and adopts a RobustScaler method to normalize the data; a long short-term memory network model and a random forest model based on historical rainfall are constructed, and model parameters are optimized in combination with a Bayesian optimization and a grid search technology; real pressure pipe water levels are systematically analyzed, potential influences of reservoir water levels on the pressure pipe water levels are filtered out, influences of accumulated period rainfall and historical rainfall on the pressure pipe water levels are compared, and a scatter diagram between the two is drawn by using a visual tool. The application is particularly suitable for small reservoir seepage data intelligent analysis, and can improve the intelligent level of reservoir safety monitoring and management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent processing and diagnosis of reservoir dam seepage pressure data, and in particular to intelligent analysis of small reservoir seepage data based on rainfall characteristics. BACKGROUND

[0002] It is known that reservoir dams, as an important part of water conservancy infrastructure, play a key role in resisting heavy rainfall, regulating water volume, and generating electricity, etc., and their safe operation during rainfall is particularly important, directly related to the social and economic stability of the downstream area. With the increase of extreme weather events, the influence of rainfall on dam seepage pressure is increasingly significant, especially during heavy or continuous rainfall, the dam body seepage pressure often appears abnormal fluctuations due to the dynamic changes of the internal seepage state of the dam.

[0003] However, the existing technology is mainly based on seepage pressure monitoring methods under static conditions. Currently, the seepage pressure threshold calculation method under the influence of rainfall mainly adopts numerical integration method and iterative calculation method, but these methods have obvious limitations in practical application. Numerical integration method usually needs to simplify the rainfall process, which is difficult to accurately capture the complex influence of dynamic changes such as rainfall intensity and frequency on dam seepage pressure. The iterative calculation method has limited accuracy when dealing with complex terrain and nonlinear material characteristics, and the response is not flexible enough under the condition of sudden rainfall. Since these methods cannot fully reflect the actual variation of seepage pressure under dynamic rainfall conditions, it is often difficult to provide accurate real-time seepage pressure prediction, reducing the reliability of dam monitoring and early warning.

[0004] In view of the above challenges, it is urgent to develop a small reservoir seepage pressure data processing and prediction method based on rainfall characteristics to more accurately capture and analyze seepage pressure data under rainfall conditions, thereby improving the accuracy of monitoring and early warning capability. SUMMARY

[0005] The present application adopts feature engineering, RobustScaler normalization, LSTM technology and random forest model to provide an efficient and accurate seepage pressure data processing scheme to meet the safety monitoring needs of reservoir dams under rainfall conditions and ensure the safe and stable operation of the dam.

[0006] The technical solution adopted by the present application to solve its technical problems is:

[0007] The intelligent analysis of small reservoir seepage data based on rainfall characteristics includes the following steps:

[0008] Step S1: Generate corresponding features for seepage pressure data, rainfall, and reservoir water level through feature construction, and use RobustScaler method to normalize these features;

[0009] Step S2: using historical cumulative rainfall data to construct input and output sequences of the LSTM model, determining the number of layers and the number of neurons in each layer, and selecting a loss function and an optimizer;

[0010] Step S3: adjusting the length of the time window, the batch size and the learning rate, optimizing the model parameters through the Bayesian network, and determining the LSTM model to predict the influence of historical rainfall on the piezometric tube water level;

[0011] Step S4: based on the cumulative rainfall in the period, constructing input features and target variables of the random forest model, selecting the number of trees and the maximum depth, and determining the loss function and the evaluation standard;

[0012] Step S5: adjusting the number of trees and the minimum sample splitting number, optimizing the parameters through grid search, and determining the random forest model to predict the influence of cumulative period rainfall on the piezometric tube water level;

[0013] Step S6: systematically analyzing the real piezometric tube water level, filtering out the potential influence of the reservoir water level on the piezometric tube water level, comparing the influence of cumulative period rainfall and historical rainfall on the piezometric tube water level, and using a visualization tool to draw a scatter plot between the two.

[0014] Further techniques of the application:

[0015] Preferably, the number of neurons in each layer in step S2 is selected to be 2 to 10 times the number of input features.

[0016] Preferably, step S3 is specifically:

[0017] Step S31, selecting the number of layers and the number of neurons in each layer of the LSTM, and determining the loss function and the optimizer of the model, so as to construct a preliminary model, and then setting the length of the time window, the batch size and the learning rate;

[0018] Step S32, starting model training using training data, optimizing hyperparameters through the Bayesian network or other optimization methods, and constantly adjusting parameters such as the length of the time window, the batch size and the learning rate to minimize the loss function; after each iteration, check the change of the loss function, if the change of the loss function is less than a preset threshold, terminate the training loop, determine the final LSTM model to predict the influence of historical rainfall on the piezometric tube water level, and obtain the final prediction result.

[0019] Preferably, step S5 uses grid search to optimize hyperparameters specifically:

[0020] The grid search traverses the possible parameter combinations, systematically evaluates the performance of the model under different parameters, and selects the best combination of the number of trees, the minimum sample split number and other hyperparameters; in this process, the number of trees and other hyperparameters are adjusted step by step through a loop; when the number of trees reaches a preset upper limit, the loop is exited, and the number of trees is stopped from being further increased.

[0021] Preferably, in step S6, a scatter plot is drawn using a visualization tool, taking the accumulated rainfall and the historical rainfall as independent variables, and taking the piezometric tube water level as the dependent variable.

[0022] The present application has the following beneficial effects:

[0023] The RobustScaler method is used for normalizing the data, effectively improving the model training efficiency; the long short-term memory network model and the random forest model based on historical rainfall are constructed, and the model parameters are optimized by combining the Bayesian optimization and the grid search technology, significantly improving the prediction accuracy; after further analyzing and filtering out the influence of the reservoir water level on the piezometric tube water level, the influence of the rainfall data on the piezometric tube water level is studied in depth; the present application is particularly suitable for intelligent analysis of small reservoir seepage data, and can improve the intelligent level of reservoir safety monitoring and management. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and all other drawings obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0025] Figure 1 Fig. 1 is a flowchart of the intelligent analysis of small reservoir seepage data based on rainfall characteristics according to the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be further described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0027] In combination with Figure 1 Fig. 1, the present application provides an intelligent analysis of small reservoir seepage data based on rainfall characteristics, which comprises:

[0028] Step S1: Generate corresponding features for osmotic pressure data, rainfall and reservoir water level through feature construction, and normalize these features using the RobustScaler method;

[0029] Step S11, divide the preprocessed data into training set and test set in proportion, take 80% of the effective data set as the training set, and the remaining 20% as the test set, then input the data into the prediction model.

[0030] Step S12, when constructing features for osmotic pressure data, rainfall and reservoir water level, generate multiple feature variables, including time series features (such as lag features, rolling window features, time period features, etc.), statistical features (mean, median, variance, etc.), cumulative features and exponentially weighted moving average features.

[0031] Step S13, apply the RobustScaler method to normalize the constructed feature variables, effectively reducing the influence of outliers on data distribution, making the data have good anti-outlier ability, and thus improving the robustness and accuracy of the model.

[0032] Step S2: Use historical cumulative rainfall data to build input and output sequences of the LSTM model, determine the number of layers and the number of neurons in each layer of the model, and select appropriate loss function and optimizer;

[0033] Step S21, use the LSTM model to capture the lagging effect of past rainfall on the current piezometric tube water level, i.e.

[0034]

[0035] where R t-1 , R t-2 , …, R t-k represent the rainfall in the past period, represents the lagging effect of past rainfall on the current piezometric tube water level. LSTM handles time series dependencies through its internal memory cells and gating mechanisms (such as forget gate, input gate and output gate), as follows:

[0036] Forget Gate: Decide which past input information needs to be "forgotten".

[0037] f t =σ(W f ·[h t-1 , R t ]+b f )

[0038] where f t is the output of the LSTM forget gate at time step t, ht-1 is the hidden state at the previous time step, W f is the weight matrix of the forget gate, R t is the input at the current time step (rainfall), b f is the bias of the forget gate.

[0039] Input Gate: controls which information needs to be updated to the memory cell.

[0040] i t = σ(W f · [h t-1 , R t ] + b i )

[0041] where i t is the output of the LSTM input gate at time step t, b i is the bias of the input gate.

[0042] Output Gate: decides which information to output from the memory cell to generate the current hidden state.

[0043] o t = σ(W f · [h t-1 , R t ] + b o )

[0044] where o t is the output of the LSTM output gate at time step t, b o is the bias of the output gate.

[0045] Step S22, determine the structure and hyperparameter configuration of the LSTM model. According to the complexity of the data, select appropriate number of LSTM layers, the number of neurons in each layer needs to be adjusted according to the task requirements, usually select 2 to 10 times of the input features, too many layers may lead to overfitting. Then, select an appropriate loss function to measure the deviation between the model's predicted value and the actual value. Finally, select the Adam optimizer to improve training efficiency and convergence speed with adaptive learning rate and momentum advantage. By reasonably configuring the model structure and hyperparameters, ensure that the LSTM model can effectively process historical rainfall data and make accurate predictions.

[0046] Step S3: adjust the time window length, batch size and learning rate, optimize the model parameters through Bayesian network to improve the accuracy of prediction;

[0047] Step S31, select the number of LSTM layers and the number of neurons in each layer, and determine the loss function and optimizer of the model, in order to build a preliminary model. Then set the key hyperparameters such as time window length, batch size and learning rate.

[0048] Step S32, start model training with training data. Optimize hyperparameters through Bayesian network or other optimization methods, constantly adjust parameters such as time window, batch size and learning rate, etc. to minimize the loss function. After each iteration, check the change of the loss function, if the change of the loss function is less than the preset threshold, terminate the training cycle, determine the final LSTM model to predict the influence of historical rainfall on the water level of the pressure pipe, and get the final prediction result.

[0049] Step S4: Based on the cumulative rainfall of the period, construct the input features and target variables of the random forest model, select the number of trees and the maximum depth, and determine the loss function and evaluation criteria;

[0050] Step S41, capture the influence of different sizes of cumulative rainfall on the water level of the pressure pipe through the random forest, and the regression prediction formula of the random forest can be expressed as:

[0051]

[0052] where f RF (R t ) is the predicted value of the random forest model under the current rainfall R t , which represents the direct influence of the current rainfall on the water level of the pressure pipe. T m (R t ) is the output of the mth decision tree based on the current rainfall R t , and M is the number of decision trees in the forest. For each decision tree, each tree learns the relationship between rainfall and pressure pipe water level, and constantly divides the data to reduce the prediction error in each leaf node. The model splits the data by rainfall R t , and the goal is to minimize the error between the residual ΔW i and the output of the tree.

[0053] Step S42, select appropriate random forest model hyperparameters according to data characteristics. First, determine the number of trees in the random forest, a larger number of trees can usually improve the stability and accuracy of the model, but will increase the computational cost. Then, set the maximum depth of each tree to prevent overfitting and ensure the generalization ability of the model. Next, select mean squared error as the loss function to measure the prediction accuracy of the model.

[0054] Step S5: Adjust the number of trees and the minimum sample split size, optimize the parameters through grid search to optimize the performance of the model and improve the prediction accuracy;

[0055] Step S51, adjust the number of trees and the minimum sample split. First, set the preliminary number of trees and the minimum sample split according to the size of the dataset and the complexity of the task. The number of trees determines the size of the forest, and a larger number of trees can improve the accuracy of the model, while the minimum sample split controls the minimum number of samples per node, avoiding overfitting and improving the generalization ability of the model. By adjusting these hyperparameters, find the appropriate combination to improve the stability and performance of the model.

[0056] Step S52, use grid search for hyperparameter optimization. By grid search, traverse the possible parameter combinations, systematically evaluate the performance of the model under different parameters, and select the best combination of tree number, minimum sample split and other hyperparameters. In this process, the number of trees and other hyperparameters are adjusted step by step through a loop. When the number of trees reaches the preset upper limit, the loop is exited and the number of trees is no longer increased. This can avoid excessive trees increasing the computational burden, while ensuring the balance between model training efficiency and prediction accuracy.

[0057] Step S6: Systematic analysis of real piezometer water level, filter out the potential influence of reservoir water level on piezometer water level. On this basis, compare the influence of period rainfall and historical rainfall on piezometer water level. Then, use the visualization tool (Matplotlib) to draw a scatter plot between the two.

[0058] Step S61, systematic analysis of real piezometer water level, first filter out the potential influence of reservoir water level on piezometer water level through SVR model. Through this process, ensure that the measured piezometer water level change can accurately reflect the influence of rainfall and other factors, thereby improving the accuracy of the data and providing a cleaner dataset for subsequent analysis.

[0059] Step S62, after filtering out the influence of reservoir water level, further analyze the influence of period rainfall and historical rainfall on piezometer water level. Use the visualization tool to draw a scatter plot, taking period rainfall and historical rainfall as independent variables and piezometer water level as dependent variable, showing the relationship between them. And through the analysis of the comparison between the piezometer water level after filtering out the influence of the reservoir water level, the cumulative rainfall and the historical rainfall on the piezometer water level, it can be clearly seen that the specific influence and accuracy of rainfall on the change of piezometer water level, which helps to better understand the correlation between the two.

[0060] The above merely describes preferred embodiments of the present application, and is not intended to limit the present application in any form; any person skilled in the art, without departing from the technical scheme of the present application, can make many possible changes and modifications to the technical scheme of the present application, or modify equivalent embodiments with equivalent changes. Therefore, any simple modification, equivalent replacement, equivalent change and modification made to the above embodiments according to the technical essence of the present application, without departing from the technical scheme of the present application, still falls within the scope of protection of the technical scheme of the present application.

Claims

1. Intelligent analysis of small reservoir seepage data based on rainfall characteristics, characterized by: The following steps are involved: Step S1: Generate corresponding features for seepage pressure data, rainfall, and reservoir water level through feature construction, and use the RobustScaler method to normalize these features; Step S2: Use historical cumulative rainfall data to construct the input and output sequences of the LSTM model, determine the number of layers of the model and the number of neurons in each layer, and select the loss function and optimizer; The LSTM model is used to capture the hysteresis effect of past rainfall on the current piezometer water level, i.e. in represents the rainfall in the past period, represents the lag effect of past rainfall on the piezometer water level at the current moment; Step S3: Adjust the time window length, batch size, and learning rate, optimize the model parameters through the Bayesian network, and determine the LSTM model to predict the impact of historical rainfall on the piezometer water level; Step S4: Based on the cumulative rainfall during this period, construct the input features and target variables of the random forest model, select the number of trees and the maximum depth, and determine the loss function and evaluation criteria; The effects of different sizes of accumulated rainfall on the piezometer water level are captured by random forest; Step S5: adjusting the number of trees and the minimum number of sample splits, optimizing the parameters through grid search, and determining the random forest model to predict the impact of cumulative rainfall on the piezometric water level; Step S6: Conduct a systematic analysis of the actual piezometer water level, filter out the potential impact of the reservoir water level on the piezometer water level, compare the impact of the cumulative rainfall during the period and the historical rainfall on the piezometer water level, and use a visualization tool to draw a scatter plot; In step S6, a visualization tool is used to draw a scatter plot, with the cumulative rainfall during the period and the historical rainfall as independent variables and the pressure pipe water level as the dependent variable.

2. The intelligent analysis of small reservoir seepage data based on rainfall characteristics according to claim 1 is characterized in that: In step S2, the number of neurons in each layer is selected to be 2 to 10 times the input features.

3. The intelligent analysis of small reservoir seepage data based on rainfall characteristics according to claim 1 is characterized in that: Step S3 specifically: Step S31: Select the number of LSTM layers and the number of neurons in each layer, and determine the loss function and optimizer of the model to build a preliminary model, and then set the time window length, batch size and learning rate; Step S32: Start model training using the training data, optimize the hyperparameters through the Bayesian network, and continuously adjust the time window, batch size, and learning rate to minimize the loss function; after each iteration, check the change of the loss function. If the change of the loss function is less than the preset threshold, terminate the training loop and determine the final LSTM model to predict the impact of historical rainfall on the pressure pipe water level and obtain the final prediction result.

4. The intelligent analysis of small reservoir seepage data based on rainfall characteristics according to claim 1 is characterized in that: In step S5, grid search is used to optimize hyperparameters. Specifically: Through grid search, we traverse possible parameter combinations, systematically evaluate the performance of the model under different parameters, and select the optimal number of trees and the minimum number of sample splits. During this process, we gradually adjust the number of trees through a loop. When the number of trees reaches the preset upper limit, we exit the loop and stop increasing the number of trees.

Images