A reservoir dam safety evaluation method based on policy optimization

Abstract

This invention relates to the field of water conservancy engineering safety monitoring technology, specifically disclosing a reservoir dam safety assessment method based on strategy optimization. The method involves real-time acquisition of sediment accumulation rate and groundwater level change rate data through a distributed sensor network, and the extraction of multi-scale features using variational mode decomposition and wavelet packet transform. These features are then combined with energy entropy and fluctuation amplitude to construct sediment anomaly feature values ​​and water level disturbance feature values. These two types of features are fused into a dam safety feature vector, which is input into a deep learning model for multi-dimensional risk assessment, outputting the dam safety level. Furthermore, a feedback control loop is constructed based on digital twin technology and a meta-learning framework to achieve dynamic adjustment of monitoring frequency and early warning thresholds, as well as automatic generation of reinforcement strategy sets, forming a closed-loop intelligent control mechanism. This improves the real-time performance, accuracy, and autonomous decision-making capability of dam safety management.

Description

Technical Field

[0001] This invention relates to the field of water conservancy project safety monitoring technology, specifically to a reservoir dam safety assessment method based on strategy optimization. Background Technology

[0002] With the continuous development of my country's water conservancy infrastructure, reservoirs and dams play a vital role in flood control, power generation, and water supply. However, due to long-term operation, changes in geological conditions, and the impact of extreme weather, dam structures may face potential risks such as unbalanced sediment distribution and abnormal seepage, thereby threatening the overall structural safety. Therefore, constructing an efficient and intelligent dam safety assessment system to achieve real-time perception and dynamic control of key risk factors has become a crucial task for ensuring the safe operation of water conservancy projects. In recent years, the integrated application of sensor monitoring, data analysis, and artificial intelligence technologies has provided a new technological path for improving the intelligence level of dam safety management.

[0003] The existing technology has the following shortcomings:

[0004] While existing research has identified dam risks using single-sensor data or empirical threshold methods, significant limitations remain in processing multi-source heterogeneous monitoring data and capturing early anomalous signals under complex operating conditions. For example, traditional methods struggle to effectively extract the nonlinear fluctuation characteristics of sedimentation processes and lack precise quantification tools for groundwater level disturbances, leading to delayed risk identification and a high false alarm rate. Furthermore, most systems remain at the stage of "static assessment – ​​manual intervention," lacking adaptive feedback mechanisms and the ability to generate reinforcement strategies based on assessment results. Summary of the Invention

[0005] The purpose of this invention is to provide a reservoir dam safety assessment method based on strategy optimization to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A reservoir dam safety assessment method based on strategy optimization includes the following steps:

[0008] S1: Real-time data collection of sediment accumulation rate and groundwater level change rate of the dam through a distributed sensor network to form a multi-source monitoring data stream;

[0009] S2: Dynamically analyze sediment accumulation rate data and calculate sediment anomaly characteristic values ​​based on their fluctuations to assess the risk of sediment distribution imbalance.

[0010] S3: Perform time-series decomposition on the groundwater level change rate data, and calculate the water level disturbance characteristic value based on the stability of its change, which is used to determine whether the seepage of the dam structure is abnormal.

[0011] S4: The sedimentation anomaly feature value and the water level disturbance feature value are fused into the dam body safety feature vector, which is then input into a pre-trained deep learning model for multi-dimensional risk assessment and outputs the dam safety level.

[0012] S5: Adaptively adjusts the monitoring frequency and early warning threshold according to the dam's safety level, generates a set of reinforcement strategies, and achieves dynamic optimization of dam safety protection through feedback control loop.

[0013] As a further aspect of the present invention: the assessment of sediment distribution imbalance risk specifically includes:

[0014] During the monitoring period, the sediment accumulation rate of the dam is monitored and the sediment accumulation rate data is collected in real time. The sediment accumulation rate data is dynamically analyzed, and based on its fluctuation, sediment anomaly characteristic values ​​are calculated. It is then determined whether the sediment anomaly characteristic values ​​are greater than or equal to a preset threshold. If so, there is a risk of sediment distribution imbalance; otherwise, there is no risk of sediment distribution imbalance.

[0015] As a further aspect of the present invention: the process for obtaining the depositional anomaly characteristic values ​​is as follows:

[0016] During the monitoring period, sediment accumulation rate data of the dam is collected in real time according to the time series to form sediment accumulation rate time series data;

[0017] The variational mode decomposition algorithm was used to perform multi-scale decomposition on the sediment accumulation rate time series data to obtain a set of intrinsic mode function components.

[0018] Calculate the energy of each intrinsic mode function separately;

[0019] Calculate the energy percentage of each intrinsic mode function;

[0020] Calculate the energy entropy of the sediment accumulation rate based on the energy proportion of each intrinsic mode function;

[0021] The maximum fluctuation amplitude of the dominant frequency component in each mode is extracted, and the energy entropy and the maximum fluctuation amplitude are constructed into a composite index in the form of a Cartesian product to calculate the sedimentation anomaly characteristic value.

[0022] As a further aspect of the present invention: the determination of whether the seepage in the dam structure is abnormal specifically includes:

[0023] During the monitoring period, the groundwater level change rate data of the dam is collected in real time according to the time series. The groundwater level change rate data is decomposed into time series. Based on the stability of its change, the water level disturbance characteristic value is calculated. It is determined whether the water level disturbance characteristic value is greater than or equal to the preset threshold. If it is, the seepage of the dam structure is abnormal. If not, the seepage of the dam structure is normal.

[0024] As a further aspect of the present invention: the process for obtaining the water level disturbance characteristic value is as follows:

[0025] During the monitoring period, the groundwater level change rate data of the dam is collected in real time according to the time series to form groundwater level change rate time series data;

[0026] Wavelet packet transform was used to perform multi-scale decomposition on the time series data of groundwater level change rate. The number of wavelet packet decomposition levels was set to [value missing]. ,get Each one carries a signal;

[0027] Calculate the normalized energy of each sub-band signal;

[0028] Calculate the wavelet packet energy entropy based on the normalized energy of each sub-band signal;

[0029] The water level disturbance characteristic value is obtained by calculating the ratio of the mean to the standard deviation of the wavelet packet energy entropy over the entire time series.

[0030] As a further aspect of the present invention: the fusion of sedimentary anomaly feature values ​​and water level disturbance feature values ​​into a dam body safety feature vector, which is then input into a pre-trained deep learning model for multi-dimensional risk assessment, specifically includes:

[0031] During the monitoring period, sedimentation anomaly characteristic values ​​and water level disturbance characteristic values ​​of the dam are acquired. These sedimentation anomaly characteristic values ​​and water level disturbance characteristic values ​​are used to construct a dam safety feature vector, which is then used as input to a deep learning model to minimize the error between the predicted dam safety score and the actual dam safety score. This serves as the training objective for the deep learning model, which is then trained. Based on the trained deep learning model, the dam safety score is output. The deep learning model is a fully connected neural network model.

[0032] As a further aspect of the present invention: the training process of the deep learning model is as follows:

[0033] The dam safety feature vector is used as the input data of the deep learning model. A fully connected neural network is selected as the model architecture. The model includes an input layer, at least two hidden layers and an output layer. The ReLU activation function is used for nonlinear mapping between each layer. The output layer uses a linear activation function to output the predicted safety score of the dam.

[0034] The objective function is to minimize the mean square error between the predicted safety score and the actual safety score.

[0035] The model parameters are iteratively updated using the stochastic gradient descent algorithm. The model parameters include: input dimension, hidden layer structure, activation function, optimization algorithm and regularization settings. After the model training is completed, the newly collected dam safety feature vectors are input into the trained deep learning model, and the current dam safety score is output.

[0036] As a further aspect of the present invention: the output dam safety level specifically includes:

[0037] If the dam's safety score is greater than or equal to a preset first threshold, the dam's safety level is high-risk; otherwise, if the dam's safety score is less than or equal to a preset second threshold, the dam's safety level is low-risk; otherwise, the dam's safety level is medium-risk.

[0038] As a further aspect of the present invention: a digital twin model is constructed based on sediment anomaly feature values ​​and water level disturbance feature values, and online fine-tuning of the model is achieved by combining a small sample task set and a meta-learning framework; the monitoring frequency and early warning threshold are dynamically adjusted through a feedback control loop, and a reinforcement strategy set is generated, the generation of which includes four steps: strategy matching, parameter optimization, effect simulation, and strategy output.

[0039] The beneficial effects of this invention are:

[0040] (1) This invention innovatively introduces variational mode decomposition (VMD) and wavelet packet transform (WPT) techniques to extract multi-scale features from non-stationary and nonlinear time-series monitoring data such as sediment accumulation rate and groundwater level change rate collected during dam operation. By performing variational mode decomposition on the time-series signal of the sedimentation process, several intrinsic mode functions (IMFs) with physical meaning are obtained, and the energy distribution and entropy value of each mode are further calculated. Combined with the fluctuation amplitude of the dominant frequency components, a composite sedimentary anomaly characteristic index is constructed. At the same time, the frequency band energy distribution characteristics of the groundwater level change rate signal are extracted using the wavelet packet multi-level decomposition method, and water level disturbance characteristic values ​​are formed based on the normalized energy entropy and its statistical characteristics (the ratio of mean to standard deviation), thereby achieving high-sensitivity identification of seepage anomalies. This method not only effectively improves the early warning capability for sedimentary imbalance and structural seepage anomalies under complex working conditions, but also significantly enhances the system's adaptability and robustness to multi-source heterogeneous monitoring data. It breaks through the limitations of traditional methods based on single threshold judgment or simple statistical analysis in dynamic risk identification, and provides more accurate and in-depth technical support for dam safety assessment.

[0041] (2) This invention deeply integrates deep learning models and digital twin technology to construct a closed-loop feedback control system with continuous learning and dynamic optimization capabilities, realizing a paradigm shift in the safety status of reservoir dams from "passive assessment" to "active protection." At the assessment level, the safety scoring model trained based on a fully connected neural network can integrate sediment anomaly feature values ​​and water level disturbance feature values ​​to obtain an accurate dam risk level classification. At the control level, by establishing a high-fidelity digital twin basic model, combining historical assessment results to generate a multi-task small sample set, and using a meta-learning framework to fine-tune the model online, the system has the ability to quickly adapt to different operating scenarios. On this basis, the system dynamically adjusts the monitoring frequency and early warning threshold according to the real-time safety level, and automatically generates a feasible reinforcement strategy scheme through four steps: strategy matching, parameter optimization, effect pre-simulation, and output execution, forming a continuous evolution mechanism of "assessment-control-feedback-optimization." This mechanism not only improves the intelligence and refinement of dam safety management, but also significantly enhances the scientific nature and timeliness of engineering decisions, and has broad promotion potential and practical application value. Attached Figure Description

[0042] The invention will now be further described with reference to the accompanying drawings.

[0043] Figure 1 This is a flowchart of a reservoir dam safety assessment method based on strategy optimization according to the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Please see Figure 1 As shown, this invention is a reservoir dam safety assessment method based on strategy optimization, comprising the following steps:

[0046] S1: Real-time data collection of sediment accumulation rate and groundwater level change rate of the dam through a distributed sensor network to form a multi-source monitoring data stream;

[0047] S2: Dynamically analyze sediment accumulation rate data and calculate sediment anomaly characteristic values ​​based on their fluctuations to assess the risk of sediment distribution imbalance.

[0048] S3: Perform time-series decomposition on the groundwater level change rate data, and calculate the water level disturbance characteristic value based on the stability of its change, which is used to determine whether the seepage of the dam structure is abnormal.

[0049] S4: The sedimentation anomaly feature value and the water level disturbance feature value are fused into the dam body safety feature vector, which is then input into a pre-trained deep learning model for multi-dimensional risk assessment and outputs the dam safety level.

[0050] S5: Adaptively adjusts the monitoring frequency and early warning threshold according to the dam's safety level, generates a set of reinforcement strategies, and achieves dynamic optimization of dam safety protection through feedback control loop.

[0051] In S1, a distributed sensor network is used to collect real-time data on sediment accumulation rate and groundwater level change rate of the dam, forming a multi-source monitoring data stream, specifically including:

[0052] In this invention, the sediment accumulation rate data and groundwater level change rate data are collected in real time through a high-precision, distributed sensing and monitoring system. This system includes multiple sensor nodes deployed in key areas of the dam. Sediment monitoring sensors are arranged in a grid pattern along the upstream and downstream riverbeds and the bottom of the reservoir to continuously measure the change in sediment deposition thickness per unit time, thereby calculating the sediment accumulation rate. Groundwater level monitoring sensors are buried in seepage observation holes at different depths inside the dam body, and combined with pressure transmitters and temperature compensation modules, to achieve high-frequency sampling of the groundwater level change rate.

[0053] All sensor nodes are connected to the central data acquisition unit via wired or wireless communication protocols, forming a unified multi-source monitoring data stream. During the acquisition process, the system synchronously acquires the raw data from each sensor at preset time intervals (e.g., once per hour), and performs preliminary filtering and outlier removal to ensure the integrity and timeliness of the data. Finally, the sediment accumulation rate data and groundwater level change rate data are stored in the database in time series form, serving as the basic input for subsequent feature extraction and safety assessment, providing reliable data support for the operation of the entire dam safety assessment system.

[0054] In S2, dynamic analysis is performed on sediment accumulation rate data. Based on its fluctuations, sedimentary anomaly characteristic values ​​are calculated to assess the risk of sediment distribution imbalance. Specifically, this includes:

[0055] During the monitoring period, the sediment accumulation rate of the dam is monitored and the sediment accumulation rate data is collected in real time. The sediment accumulation rate data is dynamically analyzed, and based on its fluctuation, sediment anomaly characteristic values ​​are calculated. It is then determined whether the sediment anomaly characteristic values ​​are greater than or equal to a preset threshold. If so, there is a risk of sediment distribution imbalance; otherwise, there is no risk of sediment distribution imbalance.

[0056] The process for obtaining the sedimentary anomaly characteristic values ​​is as follows:

[0057] During the monitoring period, sediment accumulation rate data of the dam is collected in real time according to the time series to form sediment accumulation rate time series data;

[0058] The variational mode decomposition algorithm was used to perform multi-scale decomposition on the sediment accumulation rate time series data to obtain a set of intrinsic mode function components.

[0059] Calculate the energy of each intrinsic mode function separately, using the following expression: ;

[0060] in, Indicates the first The energy of each eigenmode function This indicates the number of intrinsic mode functions. Indicates the number of sampling points. This indicates the total number of sampling points. Indicates the first The eigenmode function at the th The value of each sampling point;

[0061] The energy percentage of each intrinsic mode function is calculated using the following expression: ;

[0062] in, Indicates the first The energy percentage of each eigenmode function This represents the total number of intrinsic mode functions;

[0063] The energy entropy of the sediment accumulation rate is calculated based on the energy proportion of each intrinsic mode function. The calculation expression is as follows: ;

[0064] in, Represents energy entropy;

[0065] The maximum fluctuation amplitude of the dominant frequency component in each mode is extracted. The energy entropy and the maximum fluctuation amplitude are then combined to construct a Cartesian product-form composite index. The depositional anomaly characteristic value is calculated using the following expression: ;

[0066] in, Indicates the characteristic value of sedimentary anomalies, This indicates the maximum fluctuation range.

[0067] It should be noted that by collecting real-time sediment accumulation rate data from the dam and employing a variational mode decomposition algorithm to extract multi-scale features, the energy and proportion of each modal component are calculated, thereby constructing an energy entropy index to quantify the complexity and instability of the sedimentation process. Simultaneously, by combining the maximum fluctuation amplitude of the dominant frequency component, a Cartesian product-form composite index is formed, thus accurately calculating sedimentary anomaly characteristic values. This method not only effectively identifies the imbalance trend in sediment distribution but also overcomes the problem of poor adaptability to complex time-series signals by traditional single-threshold judgment methods, improving the sensitivity and accuracy of risk assessment. It possesses good engineering practicality and technological innovation.

[0068] In S3, the groundwater level change rate data is decomposed over time. Based on the stability of its changes, the characteristic value of water level disturbance is calculated to determine whether the seepage in the dam structure is abnormal. Specifically, this includes:

[0069] During the monitoring period, the groundwater level change rate data of the dam is collected in real time according to the time series. The groundwater level change rate data is decomposed into time series. Based on the stability of its change, the water level disturbance characteristic value is calculated. It is determined whether the water level disturbance characteristic value is greater than or equal to the preset threshold. If it is, the seepage of the dam structure is abnormal. If not, the seepage of the dam structure is normal.

[0070] The process for obtaining the water level disturbance characteristic value is as follows:

[0071] During the monitoring period, the groundwater level change rate data of the dam is collected in real time according to the time series to form groundwater level change rate time series data;

[0072] Wavelet packet transform was used to perform multi-scale decomposition on the time series data of groundwater level change rate. The number of wavelet packet decomposition levels was set to [value missing]. ,get Each one carries a signal;

[0073] The normalized energy of each sub-band signal is calculated using the following expression: ;

[0074] In the formula, Indicates the first The size carries a signal. Indicates the first Normalized energy carrying a signal, This indicates the total number of wavelet packet decomposition layers. Indicates the first One sampling point, Indicates the first The number of sampling points for each sub-band signal. Indicates the first The first in the individual belt The value of each sampling point;

[0075] Based on the normalized energy of each sub-band signal, the wavelet packet energy entropy is calculated using the following expression:

[0076] ;

[0077] In the formula, Represents the wavelet packet energy entropy;

[0078] The water level disturbance characteristic value is obtained by calculating the ratio of the mean to the standard deviation of the wavelet packet energy entropy over the entire time series.

[0079] It should be noted that by performing wavelet packet multi-scale decomposition on the groundwater level change rate data, extracting the normalized energy of each frequency band sub-band signal, and combining the wavelet packet energy entropy and its statistical characteristics (the ratio of mean to standard deviation) to calculate the water level disturbance characteristic value, the dynamic identification of the dam seepage state can be achieved. This method can effectively capture the early characteristics of abnormal groundwater level fluctuations, improve the sensitivity and stability of seepage anomaly judgment, and compared with traditional threshold methods or single frequency domain analysis methods, it has stronger anti-interference ability and engineering adaptability, providing more refined and reliable technical support for dam structural safety assessment.

[0080] In S4, sedimentary anomaly feature values ​​and water level disturbance feature values ​​are fused into a dam safety feature vector, which is then input into a pre-trained deep learning model for multi-dimensional risk assessment, outputting the dam safety level, specifically including:

[0081] During the monitoring period, sedimentation anomaly characteristic values ​​and water level disturbance characteristic values ​​of the dam are acquired. These sedimentation anomaly characteristic values ​​and water level disturbance characteristic values ​​are used to construct a dam safety feature vector, which is then used as input to a deep learning model to minimize the error between the predicted dam safety score and the actual dam safety score. This serves as the training objective for the deep learning model, which is then trained. Based on the trained deep learning model, the dam safety score is output. The deep learning model is a fully connected neural network model.

[0082] The training process of the deep learning model is as follows:

[0083] The dam safety feature vector is used as the input data of the deep learning model. A fully connected neural network is selected as the model architecture. The model includes an input layer, at least two hidden layers and an output layer. The ReLU activation function is used for nonlinear mapping between each layer. The output layer uses a linear activation function to output the predicted safety score of the dam.

[0084] The objective function is to minimize the mean square error between the predicted safety score and the actual safety score.

[0085] The model parameters are iteratively updated using the stochastic gradient descent algorithm. The model parameters include: input dimension, hidden layer structure, activation function, optimization algorithm and regularization settings. After the model training is completed, the newly collected dam safety feature vectors are input into the trained deep learning model, and the current dam safety score is output.

[0086] If the dam's safety score is greater than or equal to a preset first threshold, the dam's safety level is high-risk; otherwise, if the dam's safety score is less than or equal to a preset second threshold, the dam's safety level is low-risk; otherwise, the dam's safety level is medium-risk.

[0087] It should be noted that by fusing sedimentary anomaly feature values ​​and water level disturbance feature values ​​into a dam safety feature vector, and inputting it into a deep learning model based on a fully connected neural network, a multi-dimensional intelligent assessment of the dam's safety status is achieved. The model employs the ReLU activation function and a linear output layer structure, combined with a mean squared error objective function and a stochastic gradient descent optimization algorithm for training, which can effectively learn the nonlinear mapping relationship between complex features and safety status. By setting a dual threshold mechanism to classify the safety score output by the model, the precise classification of dam risk levels is further achieved. This method breaks through the limitations of traditional assessment methods that rely on a single indicator or empirical formula, improves the objectivity, accuracy, and interpretability of the assessment results, and provides a reliable basis for subsequent dynamic regulation, demonstrating significant technological innovation and engineering application value.

[0088] In S5, the monitoring frequency and early warning threshold are adaptively adjusted according to the dam's safety level, and a set of reinforcement strategies is generated. Dynamic optimization of dam safety protection is achieved through a feedback control loop, specifically including:

[0089] After obtaining the dam safety score and corresponding safety level output by the deep learning model, the system activates the feedback control mechanism. First, a digital twin basic model is constructed based on sedimentary anomaly characteristic values ​​and water level disturbance characteristic values ​​to simulate the current dam structure state and its potential risk evolution path. At the same time, a structured data recording module is established to associate the mapping relationship between each control action, environmental response and model prediction output.

[0090] The system utilizes historical evaluation results to construct multiple small-sample task sets, each task representing a typical dam anomaly scenario, including training and validation sets; in each task, the digital twin model is locally fine-tuned to improve its predictive ability in specific sub-scenarios; during actual operation, after each control action is executed, the system collects feedback data in real time, constructs it into a new task, inputs it into the meta-learning framework for online iteration, performs lightweight fine-tuning of the current model, and generates scenario sub-models suitable for the current working conditions;

[0091] The system periodically evaluates the predictive performance of each scenario sub-model. If the error of a sub-model on the validation set is lower than a set threshold or significantly improved compared to the main model, the parameter update information of the sub-model is extracted and fed back to the parameter set of the main model for periodic updates of the main model, or the sub-model is included in the experience model pool for reuse in similar scenarios. Based on this, the system dynamically adjusts the monitoring frequency of the sensor network and the early warning threshold for seepage anomalies according to the current dam safety level, and generates a set of reinforcement strategies by combining the simulation results of the twin model.

[0092] The process of generating the hardening strategy set includes:

[0093] Step 1: Strategy Matching: Based on the current dam safety level and dominant risk type (sedimentation imbalance or seepage anomaly), match the most suitable basic strategy template from the reinforcement strategy knowledge base;

[0094] Step 2: Parameter optimization: Based on the simulation results of the dam structure using the digital twin model, the strategy parameters, such as grouting pressure, drainage hole spacing, and reinforcement area range, are locally searched and optimized using genetic algorithms or Bayesian optimization methods.

[0095] Step 3: Effect simulation: Input the optimized reinforcement strategy into the twin model for virtual execution and simulate its impact on the dam safety score trend;

[0096] Step 4: Strategy Output: If the simulation results show that the reinforcement effect meets the expected improvement target, then add the strategy to the recommended reinforcement scheme and push it to the operation and maintenance management system for manual confirmation or automatic execution;

[0097] This forms a closed-loop evolution mechanism of "preliminary assessment - task-driven - model fine-tuning - feedback optimization - strategy generation", enabling the dam safety assessment system to continuously learn from operational data, dynamically optimize monitoring strategies, early warning mechanisms and reinforcement schemes, and achieve intelligent and adaptive control of dam safety protection.

[0098] The working principle of this invention is as follows: This invention constructs a distributed sensor network to collect data on sediment accumulation rate and groundwater level change rate in real time, forming a multi-source monitoring data stream; it uses variational mode decomposition and wavelet packet transform to extract the energy entropy and water level disturbance feature values ​​of the sedimentation process, respectively, to quantify the risks of sedimentation imbalance and seepage anomalies; it fuses the two types of features into a dam safety feature vector, inputs it into a fully connected neural network model for multi-dimensional risk assessment, and outputs the dam safety level; further, it combines digital twin modeling and meta-learning framework to dynamically adjust the monitoring frequency and early warning threshold according to the assessment results, and generates a reinforcement strategy set through four steps: strategy matching, parameter optimization, effect pre-simulation, and output execution, ultimately constructing a closed-loop system of "assessment-control-feedback-optimization" to achieve intelligent, adaptive management and dynamic protection of the dam structure's safety status.

[0099] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A reservoir dam safety assessment method based on strategy optimization, characterized in that, Includes the following steps: S1: Real-time data collection of sediment accumulation rate and groundwater level change rate of the dam through a distributed sensor network to form a multi-source monitoring data stream; S2: Dynamically analyze sediment accumulation rate data and calculate sediment anomaly characteristic values ​​based on their fluctuations to assess the risk of sediment distribution imbalance. The process for obtaining the sedimentary anomaly characteristic values ​​is as follows: During the monitoring period, sediment accumulation rate data of the dam is collected in real time according to the time series to form sediment accumulation rate time series data; The variational mode decomposition algorithm was used to perform multi-scale decomposition on the sediment accumulation rate time series data to obtain a set of intrinsic mode function components. Calculate the energy of each intrinsic mode function separately; Calculate the energy percentage of each intrinsic mode function; Calculate the energy entropy of the sediment accumulation rate based on the energy proportion of each intrinsic mode function. ; Extract the maximum fluctuation amplitude of the dominant frequency component in each mode, and calculate the depositional anomaly characteristic value based on the energy entropy and the maximum fluctuation. The calculation expression is as follows: ,in, Indicates the characteristic value of sedimentary anomalies, Indicates the maximum fluctuation range; S3: Perform time-series decomposition on the groundwater level change rate data, and calculate the water level disturbance characteristic value based on the stability of its change, which is used to determine whether the seepage of the dam structure is abnormal. The process for obtaining the water level disturbance characteristic value is as follows: During the monitoring period, the groundwater level change rate data of the dam is collected in real time according to the time series to form groundwater level change rate time series data; Wavelet packet transform was used to perform multi-scale decomposition on the time series data of groundwater level change rate. The number of wavelet packet decomposition levels was set to [value missing]. ,get Each one carries a signal; Calculate the normalized energy of each sub-band signal; Calculate the wavelet packet energy entropy based on the normalized energy of each sub-band signal; The ratio of the mean to the standard deviation of the wavelet packet energy entropy over the entire time series is calculated to obtain the characteristic value of water level disturbance. S4: The sedimentation anomaly feature value and the water level disturbance feature value are fused into the dam body safety feature vector, which is then input into a pre-trained deep learning model for multi-dimensional risk assessment and outputs the dam safety level. S5: Adaptively adjusts the monitoring frequency and early warning threshold according to the dam's safety level, generates a set of reinforcement strategies, and achieves dynamic optimization of dam safety protection through feedback control loop.

2. The reservoir dam safety assessment method based on strategy optimization according to claim 1, characterized in that, The assessment of the risk of sediment distribution imbalance specifically includes: During the monitoring period, the sediment accumulation rate of the dam is monitored and the sediment accumulation rate data is collected in real time. The sediment accumulation rate data is dynamically analyzed, and based on its fluctuation, sediment anomaly characteristic values ​​are calculated. It is then determined whether the sediment anomaly characteristic values ​​are greater than or equal to a preset threshold. If so, there is a risk of sediment distribution imbalance; otherwise, there is no risk of sediment distribution imbalance.

3. The reservoir dam safety assessment method based on strategy optimization according to claim 1, characterized in that, The determination of whether the seepage in the dam structure is abnormal specifically includes: During the monitoring period, the groundwater level change rate data of the dam is collected in real time according to the time series. The groundwater level change rate data is decomposed into time series. Based on the stability of its change, the water level disturbance characteristic value is calculated. It is determined whether the water level disturbance characteristic value is greater than or equal to the preset threshold. If it is, the seepage of the dam structure is abnormal. If not, the seepage of the dam structure is normal.

4. The reservoir dam safety assessment method based on strategy optimization according to claim 1, characterized in that, The process of fusing sedimentary anomaly feature values ​​and water level disturbance feature values ​​into a dam body safety feature vector, and inputting it into a pre-trained deep learning model for multi-dimensional risk assessment, specifically includes: During the monitoring period, sedimentation anomaly characteristic values ​​and water level disturbance characteristic values ​​of the dam are acquired. These sedimentation anomaly characteristic values ​​and water level disturbance characteristic values ​​are used to construct a dam safety feature vector, which is then used as input to a deep learning model to minimize the error between the predicted dam safety score and the actual dam safety score. This serves as the training objective for the deep learning model, which is then trained. Based on the trained deep learning model, the dam safety score is output. The deep learning model is a fully connected neural network model.

5. The reservoir dam safety assessment method based on strategy optimization according to claim 4, characterized in that, The training process of the deep learning model is as follows: The dam safety feature vector is used as the input data of the deep learning model. A fully connected neural network is selected as the model architecture. The deep learning model includes an input layer, at least two hidden layers and an output layer. The ReLU activation function is used for nonlinear mapping between each layer. The output layer uses a linear activation function to output the predicted safety score of the dam. The objective function is to minimize the mean square error between the predicted safety score and the actual safety score. The stochastic gradient descent algorithm is used to iteratively update the trainable weight parameters in the deep learning model. The network structure settings of the deep learning model include: input dimension, number of hidden layers and nodes, and activation function type. The constraints of the training process include optimization algorithm and regularization settings. After the model training is completed, the newly collected dam safety feature vector is input into the trained deep learning model, and the current dam safety score is output.

6. The reservoir dam safety assessment method based on strategy optimization according to claim 1, characterized in that, The output dam safety level specifically includes: If the dam's safety score is greater than or equal to a preset first threshold, the dam's safety level is high-risk; otherwise, if the dam's safety score is less than or equal to a preset second threshold, the dam's safety level is low-risk; otherwise, the dam's safety level is medium-risk.

7. The reservoir dam safety assessment method based on strategy optimization according to claim 1, characterized in that, The process of adaptively adjusting monitoring frequency and early warning thresholds based on the dam's safety level, generating a set of reinforcement strategies, and dynamically optimizing dam safety protection through a feedback control loop specifically includes: A digital twin model is constructed based on sediment anomaly features and water level disturbance features, and online fine-tuning of the model is achieved by combining a small sample task set and a meta-learning framework. The monitoring frequency and early warning threshold are dynamically adjusted through a feedback control loop, and a set of reinforcement strategies is generated. The generation of the reinforcement strategy set includes four steps: strategy matching, parameter optimization, effect simulation, and strategy output.

Images