Intelligent analysis and early warning method for reservoir seepage pressure data

Abstract

The invention discloses an intelligent analysis and early warning method for reservoir seepage pressure data. The method comprises the following steps: extracting a seepage pressure value corresponding to each water level value at each moment from a preliminary screening database; preliminarily judging and eliminating logic error data by combining seepage pressure measuring point pipe orifices and sensor installation elevation data; eliminating abnormal osmotic pressure data under each water level by using a 3Sigma principle; feature engineering and a Z-score method are adopted to generate features, and the features are normalized; fitting a mapping relation between the water level of the reservoir and the water level of the piezometric tube through support vector regression, calculating a model residual error and generating an early warning range; and finally, comparing and analyzing the osmotic pressure data with the early warning range, and starting a corresponding early warning mechanism according to a difference result. Based on the characteristics of reservoir dam seepage pressure monitoring data, a machine learning technology and various statistical technologies are integrated, the precision and reliability of data analysis are remarkably improved, and the method is particularly suitable for the field of reservoir dam seepage pressure data analysis and early warning requiring high precision.

Description

Technical Field

[0001] The present invention relates to an intelligent analysis and early warning method for reservoir seepage pressure data, specifically for the intelligent analysis and early warning technology of small reservoir seepage pressure monitoring data under complex working conditions. Background Art

[0002] As is well known, the dam of a reservoir, as the core of water conservancy infrastructure, not only plays a crucial role in flood control, irrigation and power generation, but also its safe operation is directly related to the social and economic stability of the downstream area. Seepage pressure monitoring data is an important indicator for evaluating the operation status of the dam. However, the existing technology mainly relies on finite element analysis and finite difference method to calculate the threshold of seepage pressure data, but this method has limitations in practical applications. First of all, finite element analysis usually relies on idealized models and assumptions, and cannot fully reflect the complexity of dam materials in the actual environment. Secondly, the finite difference method has low accuracy for complex geometries and nonlinear materials, with fixed mesh division, and it is difficult to meet the requirements of complex boundary conditions and local accuracy. In addition, the dam materials of many reservoirs are sourced locally, and their permeability coefficients are difficult to accurately determine. After years of operation and multiple reinforcement and renovation, the uniformity of the dam structure may have changed. These factors lead to a significant difference between the theoretically calculated seepage pressure threshold and the actual monitoring data, thus affecting the accuracy of seepage pressure data interpretation and posing potential risks to the safety assessment of the dam.

[0003] In view of the above challenges, there is an urgent need to develop a new method for processing and diagnosing seepage pressure data to more precisely process and analyze seepage pressure data, thereby improving the quality of monitoring data and the reliability of analysis results. Therefore, the present invention aims to provide a more efficient and accurate seepage pressure data processing solution by adopting logical error discrimination, 3Sigma principle, feature engineering, Z-score normalization and support vector regression technology, so as to better meet the needs of reservoir dam safety monitoring and ensure the safety and stability of dam operation. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent analysis and early warning method for reservoir seepage pressure data. This technology can automatically identify logical errors and eliminate abnormal data according to the data characteristics, and optimize the feature expression of the data through feature engineering and Z-score normalization processing. In addition, this technology also uses support vector regression technology for accurate fitting and prediction, and calculates the prediction residuals, thereby generating a reliable early warning range, effectively improving the accuracy and reliability of seepage pressure data analysis, and is particularly suitable for data processing and diagnosis in reservoir dam safety monitoring.

[0005] The technical solution adopted by the present invention to solve its technical problems is:

[0006] An intelligent analysis and early warning method for reservoir seepage pressure data, comprising the following steps:

[0007] Step S1: Extract the seepage pressure values corresponding to each water level value at each moment from the long-term continuous water level monitoring data of the reservoir;

[0008] Step S2: Use information such as the calibration data of the sensor to preliminarily evaluate the extracted seepage pressure values, exclude data with logical errors, and process outliers using the 3Sigma principle;

[0009] Step S3: Generate basic features and derivative features through feature engineering, and normalize these features using the Z-score normalization method;

[0010] Step S4: Based on cross-validation, select the SVR kernel function and the corresponding hyperparameter combination that perform best on the current dataset;

[0011] Step S5: Solve the optimization problem and its dual problem of the SVR model to construct a regression function for prediction;

[0012] Step S6: Evaluate the model performance through MSE ( M ean- S quare E rror, MSE) and R 2 score;

[0013] Step S7: Generate an early warning range by calculating the residuals of the SVR model and combining the standard deviation of the residuals;

[0014] Step S8: Compare and analyze the actual seepage pressure data with the early warning range, and activate the early warning mechanism according to the difference results.

[0015] Further technology of the present invention:

[0016] Preferably, step S1 is specifically:

[0017] S11. Perform data screening and cleaning operations in the long-sequence water level monitoring database of the reservoir. In particular, handle the situation where there are multiple reservoir water level values at the same moment to eliminate data redundancy and ensure data consistency. For the situation where multiple reservoir water level values are recorded at the same moment, retain the reservoir water level value that is consistent with the piezometric tube water level time. If there is no completely consistent time record, retain the reservoir water level value that is closest to the piezometric tube water level time;

[0018] S12. The unique reservoir water level value H res after screening and the piezometric tube water level value H piePerform matching to ensure that there is a unique reservoir water level corresponding to the piezometric tube water level value at each time point, providing consistent and complete input data for subsequent data analysis and diagnosis.

[0019] Preferably, step S2 is specifically as follows:

[0020] S21. For the water level of each reservoir and its corresponding seepage pressure data, calculate their means (μ 水位 , μ 渗压 ) and standard deviations (σ 水位 , σ 渗压 ). Values of the reservoir water level exceeding (μ 水位 - 3σ 水位 , μ 水位 + 3σ 水位 ), and seepage pressure values exceeding (μ 渗压 - 3σ 渗压 , μ 渗压 + 3σ 渗压 ) are marked as outliers and removed;

[0021] S22. Divide the preprocessed data into a training set and a test set and input them into the prediction model, where the training set accounts for 80% of the effective database and the test set accounts for 20% of the effective database.

[0022] Preferably, step S3 is specifically as follows:

[0023] S31. Generate basic features and derived features through feature engineering. Among them, the basic features include the reservoir water level and the piezometric tube water level, and the derived features cover time features, rolling means, difference features, lag features, and polynomial features;

[0024] S32. Normalize the features through the Z-score normalization method, convert the data of each feature into a standard normal distribution with a mean of 0 and a standard deviation of 1, so that different features are comparable. When performing Z-score normalization on feature X, the following formula will be applied to each data point in this feature:

[0025]

[0026] where X i is a data point in feature X, μ X is the mean of feature X, and σ X is the standard deviation of feature X.

[0027] Preferably, step S4 is specifically as follows:

[0028] S41. Design a cross-validation scheme, determine the number of folds and use the mean squared error and R 2As an index, define the SVR parameter search space, including the kernel function type (linear, radial basis function, polynomial) and the corresponding hyperparameter combinations (penalty parameter C, error tolerance ∈, kernel parameter);

[0029] S42. Perform cross-validation, evaluate different kernel functions and hyperparameter combinations, and select the best-performing configuration for SVR model training.

[0030] Preferably, step S5 is specifically as follows:

[0031] S51. Determine the input variables and output variables of the model;

[0032] The input variable x i is: reservoir water level, one or more features;

[0033] The output variable y i is: piezometric water level;

[0034] S52. The optimization objective of the SVR model is to find a function y that fits the data points within the ∈ tolerance range and minimizes the complexity of the model. This optimization problem is defined as:

[0035]

[0036] s.t.

[0037]

[0038] where represents the true target value of each data point in the training data; ω represents the weight vector of the model, which is the coefficient defining the decision function; b is the bias term in the model, which determines the translation of the decision function; and ξ i are slack variables, allowing the model to make deviations in some cases, so that the model can adapt to some noisy data; C represents the regularization parameter, which controls the trade-off between the fitting error on the training data and the model complexity; ∈ represents the range of tolerated errors, which defines the error size that the model can tolerate, and errors outside this range will be penalized; φ(x i ) represents the function that maps the input variable x i to a high-dimensional feature space; represents the model complexity, is the penalty term for errors outside the ∈ range, and the constraint condition means that for each data point, if the difference between the predicted value ω·φ(x i ) + b and is within ∈, then

[0039] S53. By using the Lagrange multiplier method, the original problem can be transformed into a dual problem, and the Lagrangian function of the original problem is as follows:

[0040]

[0041]

[0042] where L represents the objective function of the dual problem transformed by the Lagrange multiplier method; α, α * are Lagrange multipliers, corresponding to the slack variables ξ i and respectively. They are used for the constraint conditions and control the penalty strength of the error term; η, η * are used to control the influence of the slack variables on the objective function. The Lagrangian function of the original problem is obtained by taking the derivatives of ω, b, ξ i , setting them to 0 and simplifying to obtain the dual problem of the original problem;

[0043] S54. The final regression function is obtained by solving the dual problem of the original problem:

[0044]

[0045] where f(x) is a regression function, representing the predicted value of the final regression model obtained by solving the dual problem of the original problem. n represents the total number of training samples. K(x i , x) is a kernel function, representing the inner product of x i and x in the high-dimensional space; b is a bias term, obtained by the conditions of the support vectors.

[0046] Preferably, step S6 is specifically as follows:

[0047] S61. Calculate the MSE and R 2 scores. The specific calculation formulas are as follows:

[0048]

[0049] where n is the number of data points, is the actual value of the water level of the i-th piezometer tube, y i is the predicted value of the water level of the i-th piezometer tube, is the mean value of the actual values of the water levels of the piezometer tubes. The value range of the R 2 score is [0, 1]. The closer it is to 1, the better the fitting degree of the model to the data. MSE measures the average squared difference between the predicted value and the actual value. The closer MSE is to 0, the better the prediction effect of the model;

[0050] S62. Plot a scatter plot to visualize the difference between the actual value and the predicted value.

[0051] Preferably, step S7 is specifically as follows:

[0052] S71. Calculate the residual of each data point, that is, the difference between the actual piezometric level value and the corresponding predicted value. The specific calculation formula is as follows:

[0053]

[0054] S72. Calculate the standard deviation of the residuals, which is used to estimate the distribution of the model prediction error. The specific calculation formula is as follows:

[0055]

[0056] where σ res is the standard deviation, represents the difference between the actual piezometric level value and the corresponding predicted value, is the average value of the residuals, and n is the number of data points.

[0057] S73. Calculate the warning range. By analyzing historical abnormal data and comprehensively considering risk control and actual requirements, a 97% confidence level is selected to improve the recognition accuracy of abnormal water levels; then, the upper and lower limits of the warning range are calculated according to the confidence level. The calculation formula of the warning range is as follows:

[0058]

[0059] where is the lower limit of the warning range, that is, when the water level exceeds the corresponding warning will be activated. For a 97% confidence level, the critical value z ≈ 2.17.

[0060] Preferably, step S8 is specifically as follows:

[0061] S81. Plot the newly collected seepage pressure monitoring data on the warning range curve graph. By comparing and analyzing the actual seepage pressure data with the warning range to determine whether the data exceeds the warning range. When the actual seepage pressure data is within the warning range, the data is automatically archived and stored in the database; when the actual seepage pressure data exceeds the upper limit of the interval or is lower than the lower limit of the interval the warning mechanism will be activated.

[0062] The beneficial effects of the present invention are as follows. By strictly screening high-quality sample data, eliminating logically incorrect data, removing outliers using the 3Sigma principle, and utilizing feature engineering, Z-score normalization, and support vector regression techniques, the accuracy of data processing and diagnosis is further enhanced, making seepage pressure monitoring and prediction more accurate and reliable. The present invention is particularly applicable to the safety monitoring of key infrastructure such as reservoir dams, ensuring the efficiency and robustness of data processing, and improving the reliability and response ability of the overall monitoring system. Detailed implementation manners

[0063] In order to make the technical means, creative features, achieved purposes, and functions of the present invention easy to understand, the present invention will be further clarified below with reference to specific embodiments.

[0064] The present invention provides an intelligent analysis and early warning method for reservoir seepage pressure data, including:

[0065] Step S1: Extract the seepage pressure values corresponding to each water level value at each moment from the long-term continuous water level monitoring data of the reservoir.

[0066] S11. Perform data screening and cleaning operations in the long-sequence water level monitoring database of the reservoir. In particular, handle the situation where there are multiple reservoir water level values at the same moment to eliminate data redundancy and ensure data consistency. For the situation where multiple reservoir water level values are recorded at the same moment, retain the reservoir water level value that is consistent with the piezometric tube water level time. If there is no exactly consistent time record, retain the reservoir water level value that is closest to the piezometric tube water level time;

[0067] S12. Match the unique reservoir water level value H res after screening with the piezometric tube water level value H pie to ensure that there is a unique reservoir water level corresponding to the piezometric tube water level value at each time point, providing consistent and complete input data for subsequent data analysis and diagnosis.

[0068] Step S2: Use information such as the calibration data of the sensor to preliminarily evaluate the extracted seepage pressure values, exclude data with logical errors, and process outliers using the 3Sigma principle.

[0069] S21. For the water level of each reservoir and its corresponding seepage pressure data, calculate their means (μ 水位 , μ 渗压 ) and standard deviations (σ 水位 , σ 渗压 ). If the reservoir water level exceeds (μ 水位 - 3σ 水位 , μ 水位 + 3σ 水位 ), or the seepage pressure value exceeds (μ 渗压-3σ 渗压 、μ 渗压 +3σ 渗压 ) Values within the range are marked as outliers and removed;

[0070] S22. Divide the preprocessed data into a training set and a test set and input them into the prediction model, where the training set accounts for 80% of the effective database and the test set accounts for 20% of the effective database.

[0071] Step S3: Generate basic features and derivative features through feature engineering, and normalize these features using the Z-score normalization method.

[0072] S31. Basic features and derivative features are generated through feature engineering. Among them, the basic features include reservoir water level and piezometric tube water level, and the derivative features cover time features, rolling means, difference features, lag features, and polynomial features;

[0073] S32. Normalize the features using the Z-score normalization method, converting the data of each feature into a standard normal distribution with a mean of 0 and a standard deviation of 1, so that different features are comparable. When performing Z-score normalization on feature X, the following formula is applied to each data point in the feature:

[0074]

[0075] where, X i is a data point in feature X, μ X is the mean of feature X, and σ X is the standard deviation of feature X.

[0076] Step S4: Based on cross-validation, select the SVR kernel function and the corresponding hyperparameter combination that perform best on the current dataset.

[0077] S41. Design a cross-validation scheme, determine the number of folds and use the mean squared error and R 2 as metrics, and define the SVR parameter search space, including kernel function types (linear, radial basis function, polynomial) and the corresponding hyperparameter combinations (penalty parameter C, error tolerance ∈, kernel parameter);

[0078] S42. Perform cross-validation, evaluate different kernel functions and hyperparameter combinations, and select the best-performing configuration for SVR model training.

[0079] Step S5: Solve the optimization problem and its dual problem of the SVR model to construct a regression function for prediction.

[0080] S51. Determine the input variables and output variables of the model;

[0081] Input variable x i is: reservoir water level (one or more features);

[0082] Output variable y i is: piezometric level;

[0083] S52. The optimization objective of the SVR model is to find a function y that fits the data points within the ∈ tolerance range and minimizes the complexity of the model. This optimization problem is defined as:

[0084]

[0085] s.t.

[0086]

[0087] S52. The optimization objective of the SVR model is to find a function y that fits the data points within the ∈ tolerance range and minimizes the complexity of the model. This optimization problem is defined as:

[0088]

[0089] s.t.

[0090]

[0091] where represents the true target value of each data point in the training data; ω represents the weight vector of the model, which is the coefficient defining the decision function; b is the bias term in the model, which determines the translation of the decision function; and ξ i are slack variables that allow the model to make deviations in some cases, enabling the model to adapt to some noisy data; C represents the regularization parameter, which controls the trade-off between the fitting error on the training data and the complexity of the model; ∈ represents the range of tolerated error, which defines the size of the error that the model can tolerate, and errors outside this range will be penalized; φ(x i ) represents the function that maps the input feature x i into a high-dimensional feature space; represents the model complexity, is the penalty term for errors outside the ∈ range. The constraint means that for each data point, if the difference between the predicted value ω·φ(x i ) + b of the model and the true value is within ∈, then

[0092] S53. By the method of Lagrange multipliers, the original problem can be transformed into a dual problem. The Lagrangian function of the original problem is as follows:

[0093]

[0094] where L represents the objective function of the dual problem transformed by the Lagrange multiplier method; α, α * are Lagrange multipliers, corresponding to the slack variables ξ i and They are used for the constraint conditions and control the penalty strength of the error term; η, η * are used to control the influence of the slack variables on the objective function. The Lagrangian function of the original problem can be obtained by taking the derivatives of ω, b, ξ i , setting them to 0 and simplifying to obtain the dual problem of the original problem;

[0095] S54. Obtain the final regression function by solving the dual problem of the original problem:

[0096]

[0097] where f(x) is a regression function, representing the predicted value of the final regression model obtained by solving the dual problem of the original problem. n represents the total number of training samples, K(x i , x) is a kernel function, representing the inner product of the input data points x i and x in the high-dimensional space; b is the bias term, which can be obtained through the conditions of the support vectors.

[0098] Step S6: Evaluate the model performance through the Mean-Square Error (MSE) and R M ean- S quare E rror, MSE) and R 2 score.

[0099] S61. Calculate the MSE and R 2 score. The specific calculation formulas are as follows:

[0100]

[0101] where n is the number of data points, is the actual value of the water level of the i-th piezometer tube, y i is the predicted value of the water level of the i-th piezometer tube, is the mean value of the actual values of the water levels of the piezometer tubes. The value range of the R 2 score is [0, 1]. The closer it is to 1, the better the model fits the data. MSE measures the average squared difference between the predicted value and the actual value. The closer the MSE is to 0, the better the prediction effect of the model;

[0102] S62. Plot a scatter plot to visualize the difference between the actual values and the predicted values.

[0103] Step S7: Generate a warning range by calculating the residuals of the SVR model and combining the standard deviation of the residuals.

[0104] S71. Calculate the residual of each data point, that is, the difference between the actual piezometric level value and the corresponding predicted value. The specific calculation formula is as follows:

[0105]

[0106] S72. Calculate the standard deviation of the residuals, which is used to estimate the distribution of the model prediction error. The specific calculation formula is as follows:

[0107]

[0108] where is the average value of the residuals, and n is the number of data points;

[0109] S73. Calculate the warning range. By analyzing historical abnormal data and comprehensively considering risk control and actual needs, select a 97% confidence level to improve the recognition accuracy of abnormal water levels; then calculate the upper and lower limits of the warning range according to the confidence level. The calculation formula of the warning range is as follows:

[0110]

[0111] where for a 97% confidence level, the critical value z≈2.17.

[0112] Step S8: Compare and analyze the actual seepage pressure data with the warning range, and activate the warning mechanism according to the difference results.

[0113] S81. Plot the newly collected seepage pressure monitoring data on the warning range curve graph. By comparing and analyzing the actual seepage pressure data with the warning range to determine whether the data exceeds the warning range. When the actual seepage pressure data is within the warning range, the data is automatically archived and stored in the database; when the actual seepage pressure data exceeds the upper limit of the interval or is lower than the lower limit of the interval , activate the warning mechanism.

[0114] By comprehensively applying sample pre-screening, logical error discrimination, 3Sigma principle, feature engineering, Z-score normalization, and support vector regression technology, the ability to process and diagnose seepage pressure data has been significantly improved. This method not only enhances the accuracy and stability of data analysis, but also improves the practical application value of seepage pressure monitoring and prediction, and is especially suitable for applications in the fields of engineering safety such as reservoirs.

[0115] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention; any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make many possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above, or modify it into equivalent embodiments with equivalent changes. Therefore, any simple modification, equivalent replacement, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A reservoir seepage pressure data intelligent analysis and early warning method, characterized in that: The steps include: Step S1: extracting the seepage pressure value corresponding to each water level value at each moment from the long-term continuous water level monitoring data of the reservoir; Step S2: using the sensor calibration data and other information to make a preliminary assessment of the extracted osmotic pressure values, excluding data with logical errors, and applying the 3Sigma principle to handle abnormal values; Step S3: Generate basic features and derived features through feature engineering, and normalize these features using the Z-score standardization method; Step S4: Based on cross-validation, select the SVR kernel function and corresponding hyperparameter combination that performs best on the current data set; Step S5: solving the optimization problem of the SVR model and its dual problem, and constructing a regression function for prediction; Step S6: Through MSE and R 2 The score evaluates the model performance; Step S7: Generate a warning range by calculating the residual of the SVR model and combining the standard deviation of the residual; Step S8: Compare and analyze the actual seepage pressure data with the warning range, and activate the warning mechanism based on the difference results.

2. According to the method for intelligent analysis and early warning of reservoir seepage pressure data described in claim 1, it is characterized in that: Step S1 is specifically as follows: S11. Perform data screening and cleaning operations in the reservoir long-sequence water level monitoring database; S12, the only reservoir water level value H after screening res And the water level value of the pressure tube H pie Matching is performed to ensure that at each time point there is a unique reservoir water level and a corresponding pressure gauge water level value, providing consistent and complete input data for subsequent data analysis and diagnosis.

3. According to the method for intelligent analysis and early warning of reservoir seepage pressure data described in claim 2, it is characterized in that: In S11: Handle the situation where there are multiple reservoir water level values ​​at the same time to eliminate data redundancy and ensure data consistency. For the situation where multiple reservoir water level values ​​are recorded at the same time, retain the reservoir water level value that is consistent with the pressure gauge water level time. If there is no completely consistent time record, retain the reservoir water level value that is closest to the pressure gauge water level time.

4. According to the method for intelligent analysis and early warning of reservoir seepage pressure data described in claim 2, it is characterized in that: Step S2 is specifically as follows: S21. Calculate the mean (μ) of the water level and its corresponding seepage pressure data of each reservoir. 水位 , μ 渗压 ) and standard deviation (σ 水位 , σ 渗压 ), the reservoir water level exceeds (μ 水位 -3σ 水位 , μ 水位 +3σ 水位 ), osmotic pressure value exceeds (μ 渗压 -3σ 渗压 , μ 渗压 +3σ 渗压 ) are marked as outliers and removed; S22. Divide the preprocessed data into a training set and a test set and input them into the prediction model.

5. According to the method for intelligent analysis and early warning of reservoir seepage pressure data described in claim 4, it is characterized in that: In S22, the training set accounts for 80% of the effective database, and the test set accounts for 20% of the effective database.

Images