Water conservancy digital intelligent management system and method fused with artificial intelligence

Abstract

The invention discloses a water conservancy digital intelligent management system and method fused with artificial intelligence, belongs to the technical field of water conservancy management, and solves the problem that monitoring point selection is not comprehensive and accurate enough due to the fact that a monitoring base station is single and fixed in setting and only depends on historical fault events in an existing method. The method comprises the following steps: determining monitoring control points based on a candidate control point combination weighting result, generating a water conservancy facility map topology, identifying and analyzing a preprocessing monitoring set by an early warning management model, constructing a domino effect propagation model combined with cascade failure, and predicting cascade risk influence of the water conservancy facility map topology; according to the method, the candidate control points are comprehensively evaluated through the combined weighting method, subjective weighting and objective weighting are combined, the game theory is introduced to optimize the comprehensive weight, geographic parameters, historical monitoring information and dynamic changes of all indexes of the water conservancy facilities can be more comprehensively considered, and therefore the monitoring control points are more accurately selected; and the finally determined monitoring control point is more suitable for the actual risk distribution.

Description

Technical Field

[0001] This invention belongs to the field of water conservancy management technology, specifically relating to a water conservancy digital intelligent management system and method that integrates artificial intelligence. Background Technology

[0002] Water conservancy projects refer to the engineering field that rationally utilizes and regulates water resources through the construction and management of water conservancy facilities to meet human needs for water and promote socio-economic development. Water conservancy projects undertake core functions such as flood control and disaster reduction, water resource allocation, and water ecological protection. With the intensification of global climate change and the continuous increase in water demand due to population growth and economic development, the demand for water resources is constantly rising.

[0003] Traditional water conservancy management mainly relies on manual inspections, discrete sensor monitoring, and experience-driven decision-making, which has significant limitations: on the one hand, the existing monitoring system is mostly "single-point collection", with limited data coverage and low collection frequency, making it difficult to capture the dynamic evolution of hydrological extreme events; on the other hand, the application of water conservancy monitoring data is limited to the "storage and display" level, lacking the ability to deeply integrate and analyze multi-source heterogeneous data.

[0004] With the rapid development of artificial intelligence (AI) technology, it has demonstrated strong application potential in numerous fields. AI technology can rapidly analyze and process massive amounts of data, enabling automated and intelligent decision support, and providing new ideas and methods for solving many problems in traditional water conservancy management models. Therefore, deeply integrating AI technology with water conservancy management to construct a digital and intelligent water conservancy management system and methods is of great significance for improving water conservancy management, ensuring the safe operation of water conservancy projects, and optimizing water resource allocation.

[0005] Chinese patent CN116489687B discloses a smart water conservancy monitoring system and method based on 5G communication technology. The system includes: establishing a CPE wireless communication network to acquire facility failure events corresponding to historical maintenance records; identifying and judging the associated monitoring base stations for each facility failure event; identifying and capturing target monitoring base stations based on the event occurrence rate corresponding to each facility failure event; setting target monitoring base stations with monitoring data transmission risks as feature monitoring base stations based on the operational status data presented by each target monitoring base station; judging and judging the feature monitoring indicators of each feature monitoring base station during communication anomaly diagnosis; providing feedback to staff to monitor the network operation status of each feature monitoring base station; and adaptively adjusting the transmission mode of each feature monitoring base station based on the operational status of its feature monitoring indicators. However, existing methods identify target monitoring base stations based on historical maintenance records and event occurrence rates, setting them as feature monitoring base stations. This results in a single, fixed monitoring base station setting that relies solely on historical failure events, ignoring the complexity of the overall structure and operational status of water conservancy facilities. This leads to insufficient and inaccurate selection of monitoring points, thus affecting the monitoring and management of water conservancy facilities. To address these issues, we propose a water conservancy digital intelligent management system and method that integrates artificial intelligence. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a water conservancy digital intelligent management system and method that integrates artificial intelligence. This solves the problem that existing methods identify target monitoring base stations based on historical maintenance records and event occurrence rates, and set them as feature monitoring base stations. The monitoring base stations are set up in a single and fixed manner, and rely solely on historical fault events, ignoring the complexity of the overall structure and operating status of water conservancy facilities. This results in insufficient and inaccurate selection of monitoring points, which in turn affects the monitoring and management of water conservancy facilities.

[0007] This invention is implemented as follows: a water conservancy digital management method integrating artificial intelligence, the method comprising: Acquire basic information and historical monitoring information of water conservancy facilities, identify candidate control points in water conservancy facilities, assign weights to candidate control points using a combination weighting method, determine monitoring control points based on the combination weighting results of candidate control points, and generate a topology map of water conservancy facilities. Real-time acquisition of facility monitoring data collected from monitoring and control points in the water conservancy facility map topology; preprocessing of facility monitoring data; output of preprocessed monitoring set. A pre-built early warning management model is trained based on historical monitoring information. A pre-processed monitoring set is loaded, and the early warning management model identifies and analyzes the pre-processed monitoring set to output the dynamic balance value of the monitoring control points. The dynamic balance value of the monitoring control point is compared and analyzed with the preset risk threshold to determine whether the monitoring control point is abnormal. If the monitoring control point is determined to be an abnormal monitoring point, an abnormal monitoring point command is triggered. Obtain at least one set of abnormal monitoring points, construct a domino effect propagation model based on the abnormal monitoring points and combine cascading failures, predict the cascading risk impact of the water conservancy facility map topology, and provide feedback on the cascading risk impact of the water conservancy facility map topology.

[0008] Preferably, the method for assigning combined weights to candidate control points using a combined weighting method includes: Acquire basic information on water conservancy facilities, identify the geographical parameters of the facilities in the basic information, and generate candidate control points based on the geographical parameters and historical monitoring information for monitoring the structural safety, seepage safety, stress-strain safety, operational status, environmental impact, water use efficiency, soil moisture, drainage network, and bank slope stability of water conservancy facilities. Using structural safety, seepage safety, and risk correlation as indicator layers, the candidate control points are subjectively weighted based on the analytic hierarchy process. A subjective weight matrix is ​​constructed through pairwise comparisons, and a consistency test is used to obtain the subjective weights of the candidate control points. The coefficient of variation of the influence indicators associated with candidate control points is determined based on the coefficient of variation method. Objective weights are then assigned to the candidate control points based on the coefficient of variation of the influence indicators associated with them. The subjective and objective weights of candidate control points are loaded, and the comprehensive weight including subjective and objective weights is determined based on game theory. When calculating the comprehensive weight, the subjective and objective weights of candidate control points are linearly combined, and the coefficients of the linear combination are optimized based on game theory to calculate the optimal comprehensive weight, thereby obtaining the comprehensive weight of candidate control points. The combined weighting result including the comprehensive weight is output. Iterate through the combined weighting results and determine whether the comprehensive weight of the candidate control points exceeds the preset weighting threshold; If the overall weight of a candidate control point exceeds the preset weighting threshold, the candidate control point will be set as a monitoring control point. If the overall weight of the candidate control points does not exceed the preset weighting threshold, the candidate control points are merged based on their objective weights and spatiotemporal correlation, and the merged control point result is set as the monitoring control point. Load at least one set of monitoring and control points and identify the facility geographic parameters of the monitoring and control points. Use a 3D modeling artifact to construct a topology map of water conservancy facilities containing the monitoring and control points.

[0009] Preferably, the method for merging candidate control points by combining objective weights and spatiotemporal correlation includes: The objective weights of candidate control points are identified, the coefficient of variation of candidate control points is extracted by the back-inference method, and the control point information entropy, which represents the coefficient of variation of candidate control points and the comprehensive information of facility geographical parameters, is calculated based on the normal distribution. Load the control point information entropy of the candidate control point, determine whether the control point information entropy exceeds the preset spatiotemporal information threshold, and if the control point information entropy exceeds the preset spatiotemporal information threshold, set the candidate control point as the center control point. Based on the control point information entropy of the central control point, a control point spatial weight matrix and a time correlation coefficient matrix are constructed, where the spatial weight matrix is ​​a distance decay matrix and the time correlation coefficient matrix is ​​a Pearson correlation coefficient matrix. Multiply the spatial weight matrix of the control points and the temporal correlation coefficient matrix to obtain the comprehensive spatiotemporal correlation matrix. Analyze the comprehensive spatiotemporal correlation matrix and select 40% of the quantile of the comprehensive spatiotemporal correlation matrix as the mutual information threshold. Based on spatiotemporal correlation analysis, the spatiotemporal correlation mutual information between the central control point and the associated control points is calculated. It is then determined whether the spatiotemporal correlation mutual information between the central control point and the associated control points is greater than the mutual information threshold. If the spatiotemporal correlation mutual information between the central control point and the associated control points is greater than the mutual information threshold, the associated control points and the central control point are merged, and the merged control point result is set as the monitoring control point.

[0010] Preferably, the method for preprocessing facility monitoring data includes: Acquire facility monitoring data, remove outliers from the facility monitoring data, and output the facility monitoring data after outlier removal. Perform spectrum analysis on the facility monitoring data after outlier processing, and adaptively decompose the facility monitoring data based on the spectrum analysis results to obtain at least one set of monitoring signal segments; The monitoring data of facilities within the monitoring signal segment is converted into the frequency domain to obtain the signal spectrum. The energy distribution of the signal spectrum is calculated, and the energy value corresponding to the noise in the signal spectrum is output. Load at least one set of signal spectrograms, identify the noise corresponding to the signal spectrograms, and use wavelet soft threshold denoising to remove the noise corresponding to the signal spectrograms to generate a preprocessed monitoring set.

[0011] Preferably, the method for training a pre-built early warning management model based on historical monitoring information includes: Load a pre-built early warning management model, and preset the hyperparameters, training rounds, loss function, and activation function of the early warning management model; Historical monitoring information is acquired, and the Laplace mechanism is used to add noise to the historical monitoring information. The noise-added historical monitoring information is then divided into a training set and a test set. The early warning management model is iteratively trained using a training set. During iterative training, Bayesian optimization is used to adjust the model hyperparameters, and the output of each layer is calculated. The weights and biases are adjusted to minimize the prediction error and output a converged early warning management model. The test set is acquired, and the early warning management model identifies and analyzes the test set, outputs the test balance value, identifies the impact indicators associated with the test balance value, and determines whether the impact indicators associated with the test balance value are consistent with the actual impact indicators. If the impact indicators associated with the test balance value are consistent with the actual impact indicators, the converged early warning management model is output. If the impact indicators associated with the test equilibrium value are inconsistent with the actual impact indicators, the SGD optimizer is used to update and optimize the model hyperparameters, and the early warning management model is iteratively trained.

[0012] Preferably, the early warning management model uses a convolutional neural network model as the initial model. The initial model includes an input layer, an output layer, and a convolutional network architecture. The input layer introduces the PBFT consensus algorithm, which integrates artificial intelligence algorithms, and a weighted bidirectional feature pyramid network. The convolutional network architecture includes convolutional layers, pooling layers, and fully connected layers. A long short-term memory neural network is added after the fully connected layers. The DPoS consensus algorithm is introduced into the long short-term memory neural network. A TCN module based on fuzzy hierarchical analysis is added between the convolutional network architecture and the output layer. The TCN module uses fuzzy hierarchical analysis combined with triangular fuzzy number theory to fuzzily couple the anomaly degree and risk degree of the monitoring points in the preprocessed monitoring set, and outputs the dynamic balance value of the monitoring control points.

[0013] Preferably, the early warning management model's method for identifying and analyzing the preprocessed monitoring set includes: The preprocessed monitoring set is obtained, and the input layer precodes the facility monitoring data in the preprocessed monitoring set based on the PBFT consensus algorithm. The weighted bidirectional feature pyramid network fuses pre-encoded facility monitoring data in a bottom-up manner and outputs the data fusion features corresponding to the monitoring control points. The comprehensive weight of the monitoring and control points is assigned to the data fusion feature of the monitoring and control points. The convolutional network architecture extracts at least one set of local anomaly features from the data fusion feature, and performs nonlinear activation on the local anomaly features to obtain the activated anomaly feature map. A weighted adjacency matrix is ​​then constructed based on the activated anomaly feature map. The fully connected layer uses the Leiden algorithm to calculate the weighted adjacency matrix to obtain the anomaly degree of the monitoring points; The monitoring point anomaly score is loaded, and the long short-term memory neural network predicts and assesses the risk of the monitoring control point based on the monitoring point anomaly score and data fusion features, and outputs the monitoring point risk score corresponding to the monitoring control point. The TCN module uses fuzzy hierarchical analysis combined with triangular fuzzy number theory to fuzzily couple the anomaly degree and risk degree of monitoring points in the preprocessed monitoring set, and outputs the dynamic balance value of the monitoring control points.

[0014] Preferably, the method for constructing a domino effect propagation model based on anomaly monitoring points and incorporating cascading failures includes: Obtain at least one set of abnormal monitoring points, construct an undirected graph of risk propagation by combining the facility geographic parameters corresponding to the abnormal monitoring points, and add the dynamic balance value of the abnormal monitoring points to the undirected graph of risk propagation. In the undirected graph of risk propagation, the graph nodes represent abnormal monitoring points, and the weight of the graph connecting edges represents the risk propagation intensity. A Bayesian network probability model is used to calculate the probability of risk propagation between nodes, and a time decay factor is introduced into the undirected graph of risk propagation. The failure of risk propagation is quantified by Markov logic network. By loading a Markov logic network and an undirected graph of risk propagation, and fusing the Markov logic network and the undirected graph of risk propagation, a domino effect propagation model is obtained. The impact indicators of water conservancy accidents are extracted using a domino effect propagation model, and the cascading risk impact of the water conservancy facility topology is extracted based on a cascading failure algorithm, thus providing feedback on the cascading risk impact of the water conservancy facility topology.

[0015] On the other hand, the present invention also provides a water conservancy digital intelligence management system that integrates artificial intelligence, the water conservancy digital intelligence management system integrating artificial intelligence includes: The graph topology module is used to acquire basic information and historical monitoring information of water conservancy facilities, identify candidate control points in water conservancy facilities, assign weights to candidate control points through a combination weighting method, determine monitoring control points based on the combination weighting results of candidate control points, and generate a graph topology of water conservancy facilities. The monitoring data acquisition module is used to acquire facility monitoring data collected from monitoring control points in the water conservancy facility map topology in real time, preprocess the facility monitoring data, and output a preprocessed monitoring set. The monitoring and analysis module trains a pre-built early warning management model based on historical monitoring information, loads a pre-processed monitoring set, identifies and analyzes the pre-processed monitoring set, outputs the dynamic balance value of the monitoring control point, and compares and analyzes the dynamic balance value of the monitoring control point with the preset risk threshold to determine whether the monitoring control point is abnormal. If the monitoring control point is determined to be an abnormal monitoring point, an abnormal monitoring point command is triggered. The cascade analysis module is used to acquire at least one set of abnormal monitoring points, construct a domino effect propagation model based on the abnormal monitoring points and cascade failures, predict the cascade risk impact of the water conservancy facility map topology, and provide feedback on the cascade risk impact of the water conservancy facility map topology.

[0016] Preferably, the graph topology module includes: The control point candidate unit is used to obtain basic information of water conservancy facilities, identify the geographical parameters of the facilities in the basic information of water conservancy facilities, and generate candidate control points based on the geographical parameters of the facilities and historical monitoring information for monitoring the impact indicators of structural safety, seepage safety, stress-strain safety, operating status, environmental correlation, water use efficiency, soil moisture, drainage network and bank slope stability of water conservancy facilities. The subjective weighting unit uses structural safety, seepage safety, and risk correlation as indicator layers. It subjectively weights candidate control points based on the analytic hierarchy process, constructs a subjective weight matrix through pairwise comparisons, and obtains the subjective weights of candidate control points using a consistency test. The objective weighting unit determines the coefficient of variation of the influence indicators associated with the candidate control points based on the coefficient of variation method, and performs objective weighting based on the coefficient of variation of the influence indicators associated with the candidate control points to obtain the objective weight of the candidate control points. The comprehensive weighting unit loads the subjective and objective weights of the candidate control points, determines the comprehensive weight including subjective and objective weights based on game theory, and calculates the comprehensive weight including subjective and objective weights by linearly combining the subjective and objective weights of the candidate control points during the comprehensive weight calculation. The coefficients of the linear combination are optimized based on game theory to calculate the optimal comprehensive weight, obtain the comprehensive weight of the candidate control points, and output the combined weighting result including the comprehensive weight. The control point determination unit is used to traverse the combined weighting results and determine whether the comprehensive weight of the candidate control point exceeds the preset weighting threshold. If the comprehensive weight of the candidate control point exceeds the preset weighting threshold, the candidate control point is set as a monitoring control point. If the comprehensive weight of the candidate control point does not exceed the preset weighting threshold, the candidate control points are merged by combining the objective weight and spatiotemporal correlation of the candidate control points, and the merged control point result is set as a monitoring control point. The graph topology construction unit is used to load at least one set of monitoring and control points and identify the facility geographic parameters of the monitoring and control points. It uses a 3D modeling artifact to construct a water conservancy facility graph topology containing the monitoring and control points.

[0017] Compared with the prior art, the embodiments of this application have the following main advantages: In this embodiment of the invention, when determining monitoring control points and generating a water conservancy facility map topology based on the combined weighting results of candidate control points, a comprehensive evaluation of candidate control points is conducted through a combined weighting method. By combining subjective and objective weighting and introducing game theory to optimize the comprehensive weight, the geographical parameters, historical monitoring information, and dynamic changes of various indicators of water conservancy facilities can be considered more comprehensively. This allows for more accurate selection of monitoring control points, making the final determined monitoring control points more consistent with the actual risk distribution and covering key risk areas. This significantly improves the intelligence level and scientific decision-making of water conservancy digital management.

[0018] In this embodiment of the invention, a combined weighting method is used to assign weights to candidate control points. When a candidate control point is precisely designated as a monitoring control point, subjective and objective weights are linearly combined, and the linear combination coefficients are optimized using game theory to calculate the optimal comprehensive weight. This avoids the bias that may be caused by a single weighting method. By judging the comprehensive weight of candidate control points through a preset weighting threshold, monitoring control points can be selected dynamically. Furthermore, the threshold can be flexibly adjusted according to actual needs and risk levels to ensure more reasonable selection of monitoring points. For candidate control points whose comprehensive weight does not exceed the threshold, they are merged by combining objective weights and spatiotemporal correlation, further optimizing the distribution of monitoring points. This merging strategy not only reduces the number of monitoring points and lowers monitoring costs, but also improves the representativeness of monitoring through merging.

[0019] This invention provides a method for merging candidate control points by combining objective weights and spatiotemporal correlation. The method extracts the coefficient of variation of candidate control points using a back-calculation method and calculates the information entropy of control points based on a normal distribution. This allows for a scientific evaluation of the objective weight of each candidate control point. By determining the information entropy of control points using preset spatiotemporal information thresholds, the method can dynamically identify the central control point. This enables dynamic adjustment of the threshold based on actual spatiotemporal data, ensuring accurate positioning of the central control point. The comprehensive spatiotemporal correlation analysis comprehensively assesses the relationships between control points, considering not only spatial proximity but also temporal correlation. This optimizes the efficiency of monitoring resource allocation and provides a more reliable monitoring and control point foundation for risk early warning and prevention of water conservancy facilities.

[0020] In this embodiment of the invention, the early warning management model uses a convolutional neural network model as the initial model. The initial model incorporates the PBFT consensus algorithm, which integrates artificial intelligence algorithms, and a weighted bidirectional feature pyramid network. This overcomes data conflicts caused by equipment errors and transmission delays from different sensors in water conservancy monitoring. A long short-term memory neural network is added after the fully connected layer, enabling simultaneous processing of spatial and temporal features, making it suitable for the complex characteristics of water conservancy facility monitoring data. The DPoS consensus algorithm is introduced into the long short-term memory neural network, and the combination of fuzzy hierarchical analysis and the TCN module integrates expert experience with data-driven approaches, outputting dynamic equilibrium values ​​that better meet engineering requirements. Furthermore, the integration of triangular fuzzy number theory further enhances the model's ability to handle uncertainty. This allows for more accurate description and processing of fuzzy information in monitoring data, improving the model's robustness and significantly reducing the false alarm rate. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the implementation process of the intelligent water conservancy management method integrating artificial intelligence provided by the present invention.

[0022] Figure 2The diagram illustrates the process of assigning combined weights to candidate control points using a combined weighting method.

[0023] Figure 3 A schematic diagram of the process for merging candidate control points by combining objective weights and spatiotemporal correlation is shown.

[0024] Figure 4 A schematic diagram of the implementation process of the facility monitoring data preprocessing method is shown.

[0025] Figure 5 A schematic diagram illustrating the implementation process of a pre-built early warning management model based on historical monitoring information is shown.

[0026] Figure 6 The diagram illustrates the implementation process of the preprocessing monitoring set identification and analysis method in the early warning management model.

[0027] Figure 7 A schematic diagram illustrating the implementation process of a method for constructing a domino effect propagation model based on anomaly monitoring points and cascading failures is shown.

[0028] Figure 8 A schematic diagram of the structure of a water conservancy digital management system that integrates artificial intelligence is shown. Detailed Implementation

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0030] Existing methods identify target monitoring base stations based on historical maintenance records and event occurrence rates, designating them as characteristic monitoring base stations. This approach relies on a single, fixed base station configuration and depends solely on historical fault events, neglecting the complexity of the overall structure and operational status of water conservancy facilities. This results in incomplete and inaccurate selection of monitoring points, impacting the monitoring and management of water conservancy facilities. To address these issues, we propose a water conservancy digital intelligent management system and method integrating artificial intelligence. In short, the method first acquires basic information and historical monitoring information of water conservancy facilities, identifies candidate control points within the facilities, assigns weights to these candidate control points using a combined weighting method, acquires facility monitoring data collected from monitoring control points in the water conservancy facility topology in real time, preprocesses the monitoring data, outputs a preprocessed monitoring set, and an early warning management model analyzes and identifies the preprocessed monitoring set, outputting dynamic balance values ​​for the monitoring control points to determine if they are abnormal. If a monitoring control point is identified as an abnormal monitoring point, a domino effect propagation model combining cascading failures is finally constructed based on the abnormal monitoring point to predict the cascading risk impact of the water conservancy facility topology and provide feedback on the cascading risk impact of the water conservancy facility topology. In this embodiment of the invention, when determining monitoring control points and generating a water conservancy facility map topology based on the combined weighting results of candidate control points, a comprehensive evaluation of candidate control points is conducted through a combined weighting method. By combining subjective and objective weighting and introducing game theory to optimize the comprehensive weight, the geographical parameters, historical monitoring information, and dynamic changes of various indicators of water conservancy facilities can be considered more comprehensively. This allows for more accurate selection of monitoring control points, making the final determined monitoring control points more consistent with the actual risk distribution and covering key risk areas. This significantly improves the intelligence level and scientific decision-making of water conservancy digital management.

[0031] This invention provides a water conservancy digital management method that integrates artificial intelligence. Figure 1 A schematic diagram illustrating the implementation process of a water conservancy digital management method integrating artificial intelligence is shown. This method includes: S10: Obtain basic information and historical monitoring information of water conservancy facilities, identify candidate control points in water conservancy facilities, assign weights to candidate control points using a combination weighting method, determine monitoring control points based on the combination weighting results of candidate control points, and generate a topology map of water conservancy facilities. It should be noted that the water conservancy facilities mentioned include, but are not limited to, reservoir dams, dike projects, irrigation projects, hydropower stations, and drainage pipe networks. The basic information of water conservancy facilities includes, but is not limited to, the geographical parameters of the facilities, the engineering facility parameters (reservoir capacity, type, and size; number of sluice gates, gate type, and maximum flood discharge; design head and rated power of pumping stations; length and type of drainage pipe networks), facility operation and management information, hydrological and meteorological information, and environmental and ecological information. The historical monitoring information includes, but is not limited to, structural safety monitoring information, leakage safety monitoring information, strain and stress monitoring information, operational status monitoring information, environmental correlation monitoring information, soil moisture monitoring information, crop water requirement monitoring information, bank slope stability monitoring information, and river channel evolution monitoring information.

[0032] S20: Real-time acquisition of facility monitoring data collected from monitoring and control points in the water conservancy facility map topology; preprocessing of facility monitoring data; output of preprocessed monitoring set. It should be noted that the monitoring data collected by the monitoring and control points are periodically collected using a variety of sensors, including but not limited to vibrating wire displacement gauges, hydrostatic levels, total stations, crack gauges, ultrasonic water level gauges, gate opening gauges, temperature sensors, acoustic Doppler current profilers (ADCP), sediment monitors, online water quality analyzers, ground penetrating radar (GPR), infrared thermal imagers, drone inspections, ecological flow monitoring devices, fish behavior observation cameras, and water temperature sensors.

[0033] S30: Based on historical monitoring information, a pre-constructed early warning management model is trained, a pre-processed monitoring set is loaded, the early warning management model identifies and analyzes the pre-processed monitoring set, and outputs the dynamic balance value of the monitoring control points. S40, compare and analyze the dynamic balance value of the monitoring control point with the preset risk threshold to determine whether the monitoring control point is abnormal; In this embodiment of the invention, the risk threshold can be set using principal component analysis or expert consultation. The risk threshold can be set to 0.6. If the dynamic balance value of the monitoring and control point exceeds the risk threshold, the corresponding monitoring and control point is judged to be abnormal.

[0034] S50, if the monitoring control point is determined to be an abnormal monitoring point, trigger the monitoring point abnormal command; If the monitoring control point is determined to be a non-abnormal monitoring point, the dynamic balance value of the monitoring control point is fed back. S60, acquire at least one set of abnormal monitoring points, construct a domino effect propagation model combining cascading failures based on the abnormal monitoring points, predict the cascading risk impact of the water conservancy facility map topology, and provide feedback on the cascading risk impact of the water conservancy facility map topology.

[0035] In this embodiment of the invention, when determining monitoring control points and generating a water conservancy facility map topology based on the combined weighting results of candidate control points, a comprehensive evaluation of candidate control points is conducted through a combined weighting method. By combining subjective and objective weighting and introducing game theory to optimize the comprehensive weight, the geographical parameters, historical monitoring information, and dynamic changes of various indicators of water conservancy facilities can be considered more comprehensively. This allows for more accurate selection of monitoring control points, making the final determined monitoring control points more consistent with the actual risk distribution and covering key risk areas. This significantly improves the intelligence level and scientific decision-making of water conservancy digital management.

[0036] This invention provides a method for assigning combined weights to candidate control points using a combined weighting method. Figure 2 The diagram illustrates a process for assigning combined weights to candidate control points using a combined weighting method. This method includes: S101: Acquire basic information on water conservancy facilities, identify the geographical parameters of the facilities in the basic information, and generate candidate control points based on the geographical parameters and historical monitoring information for monitoring the structural safety, seepage safety, stress-strain safety, operating status, environmental correlation, water use efficiency, soil moisture, drainage network, and bank slope stability of water conservancy facilities. This comprehensiveness ensures that no important monitoring points are missed in subsequent analysis, thereby improving the integrity and reliability of the entire water conservancy facility monitoring system. S102 uses structural safety, seepage safety, and risk correlation as indicator layers. It subjectively assigns weights to candidate control points based on the analytic hierarchy process, constructs a subjective weight matrix through pairwise comparisons, and uses consistency checks to obtain the subjective weights of candidate control points. It should be noted that by constructing a subjective weight matrix and conducting consistency checks, the scientific validity and rationality of the subjective weighting were ensured. This allows for full consideration of expert experience and knowledge, making the weighting process more targeted and practical.

[0037] S103, Based on the coefficient of variation method, determine the coefficient of variation of the influence indicators associated with the candidate control points, and objectively assign weights to the candidate control points according to the coefficient of variation of the influence indicators associated with the candidate control points to obtain the objective weights of the candidate control points. S104, Load the subjective and objective weights of the candidate control points, determine the comprehensive weight including subjective and objective weights based on game theory, wherein, when calculating the comprehensive weight, the subjective and objective weights of the candidate control points are linearly combined, the coefficients of the linear combination are optimized based on game theory, the optimal comprehensive weight is calculated, the comprehensive weight of the candidate control points is obtained, and the combined weighting result including the comprehensive weight is output. S105, Traverse the combined weighting results and determine whether the comprehensive weight of the candidate control points exceeds the preset weighting threshold, wherein the weighting threshold can be 0.2-0.8; S106, If the overall weight of the candidate control point exceeds the preset weighting threshold, the candidate control point is set as a monitoring control point; S107, If the comprehensive weight of the candidate control points does not exceed the preset weighting threshold, the candidate control points are merged by combining the objective weight and spatiotemporal correlation of the candidate control points, and the merged control point result is set as the monitoring control point; S108, load at least one set of monitoring and control points and identify the facility geographic parameters of the monitoring and control points, and use a 3D modeling artifact to construct a topology map of water conservancy facilities containing the monitoring and control points.

[0038] In this embodiment of the invention, a combined weighting method is used to assign weights to candidate control points. When a candidate control point is precisely designated as a monitoring control point, subjective and objective weights are linearly combined, and the linear combination coefficients are optimized using game theory to calculate the optimal comprehensive weight. This avoids the bias that may be caused by a single weighting method. By judging the comprehensive weight of candidate control points through a preset weighting threshold, monitoring control points can be selected dynamically. Furthermore, the threshold can be flexibly adjusted according to actual needs and risk levels to ensure more reasonable selection of monitoring points. For candidate control points whose comprehensive weight does not exceed the threshold, they are merged by combining objective weights and spatiotemporal correlation, further optimizing the distribution of monitoring points. This merging strategy not only reduces the number of monitoring points and lowers monitoring costs, but also improves the representativeness of monitoring through merging.

[0039] This invention provides a method for merging candidate control points by combining objective weights and spatiotemporal correlation. Figure 3 The diagram illustrates the implementation flow of a method for merging candidate control points by combining objective weights and spatiotemporal correlation. This method includes: S201 identifies the objective weights of candidate control points, uses a reverse calculation method to extract the coefficient of variation of candidate control points, calculates the control point information entropy based on the coefficient of variation of candidate control points and the comprehensive information of facility geographical parameters based on the normal distribution, and calculates the information entropy of control points based on the normal distribution and associated control point information. This enables the accurate measurement of the "information content" of control points in terms of spatial distribution and attribute characteristics, so that the control point information entropy can comprehensively consider data sensitivity and spatial representativeness, avoiding "missed detection" or "misselection" due to the bias of a single indicator. The control point information entropy is calculated using the following formula: (1) in, Represents the control point information entropy. This represents the coefficient of variation for the candidate control points. This represents the associated control points derived from candidate control points extracted based on facility geographic parameters. This indicates the impact indicators included in the candidate control points; S202, Load the control point information entropy of the candidate control points, and determine whether the control point information entropy exceeds the preset spatiotemporal information threshold. If the control point information entropy exceeds the preset spatiotemporal information threshold, set the candidate control point as the central control point. It should be noted that control points with information entropy exceeding the threshold are usually located in critical areas of the facility (such as dam shoulders, areas with high leakage rates) or are sensitive to multi-dimensional safety indicators. Therefore, the corresponding candidate control points are used as the merged "core anchor points", that is, the central control points. Information entropy is used to measure the uncertainty or amount of information in the attribute values ​​of control points. The higher the entropy value, the more complex and unpredictable the attribute changes of the control point are, which usually means that it has a more critical position in the system. S203, based on the control point information entropy of the central control point, construct the control point spatial weight matrix and the time correlation coefficient matrix. The spatial weight matrix is ​​a distance decay matrix, and the time correlation coefficient matrix is ​​a Pearson correlation coefficient matrix. The spatial weight matrix can accurately reflect the geographical proximity relationship between control points, while the comprehensive spatiotemporal correlation matrix can integrate spatial and temporal dimensions, thereby comprehensively reflecting the spatiotemporal coupling relationship between control points. S204. Multiply the spatial weight matrix of the control points and the temporal correlation coefficient matrix to obtain the comprehensive spatiotemporal correlation matrix. Analyze the comprehensive spatiotemporal correlation matrix and select 40% of the quantile of the comprehensive spatiotemporal correlation matrix as the mutual information threshold. S205, calculate the spatiotemporal correlation mutual information between the central control point and the associated control points based on spatiotemporal correlation analysis; S206, Determine whether the spatiotemporal correlation mutual information between the central control point and the associated control point is greater than the mutual information threshold; S207, If the spatiotemporal correlation mutual information between the central control point and the associated control point is greater than the mutual information threshold, the associated control point and the central control point are merged, and the merged control point result is set as the monitoring control point; The spatiotemporal correlation mutual information between the central control point and associated control points is calculated using the following formula: (2) (3) (4) (5) in, This represents the spatiotemporal correlation mutual information between the central control point and associated control points. These represent the spatial correlation coefficient and the temporal-spatial coefficient between the central control point and the associated control points, respectively. These represent the spatial weight matrix and the temporal correlation coefficient matrix of the control points, respectively. To synthesize the spatiotemporal correlation matrix, Represents the spacetime balance coefficient. These are the spatial attribute values ​​of the central control point and the associated control points, respectively. These are the attribute averages of the central control point and the associated control points, respectively.

[0040] This invention provides a method for merging candidate control points by combining objective weights and spatiotemporal correlation. The method extracts the coefficient of variation of candidate control points using a back-calculation method and calculates the information entropy of control points based on a normal distribution. This allows for a scientific evaluation of the objective weight of each candidate control point. By determining the information entropy of control points using preset spatiotemporal information thresholds, the method can dynamically identify the central control point. This enables dynamic adjustment of the threshold based on actual spatiotemporal data, ensuring accurate positioning of the central control point. The comprehensive spatiotemporal correlation analysis comprehensively assesses the relationships between control points, considering not only spatial proximity but also temporal correlation. This optimizes the efficiency of monitoring resource allocation and provides a more reliable monitoring and control point foundation for risk early warning and prevention of water conservancy facilities.

[0041] This invention provides a method for preprocessing facility monitoring data. Figure 4 A schematic diagram of the implementation process of a facility monitoring data preprocessing method is shown. The facility monitoring data preprocessing method includes: S301, Obtain facility monitoring data. Considering that outliers will increase the amount of data and computational complexity, the outliers in the facility monitoring data are deleted, and the outlier-processed facility monitoring data is output. S302, perform spectrum analysis on the facility monitoring data after outlier processing and adaptively decompose the facility monitoring data based on the spectrum analysis results to obtain at least one set of monitoring signal segments. In this embodiment of the invention, the spectrum analysis of the facility monitoring data after outlier processing can be performed using Fourier transform processing. The spectrum results include the main frequency and energy ratio. When adaptively decomposing the facility monitoring data based on the spectrum analysis results, the empirical mode decomposition (EMD) method can be used for decomposition processing. S303, frequency domain conversion is performed on the facility monitoring data within the monitoring signal segment to obtain the signal spectrum, the energy distribution of the signal spectrum is calculated, and the energy value corresponding to the noise in the signal spectrum is output. In this embodiment of the invention, considering that the frequency characteristics of different noise sources in a multi-source heterogeneous manner are different, the energy distribution calculation is used to clarify the distribution range of the noise, thereby avoiding the accidental deletion of effective signals during noise reduction. The noise energy value can be used to assess the degree of influence of noise on the data, thereby facilitating the wavelet soft threshold noise reduction method to perform noise reduction processing for noise with different degrees of influence. S304, load at least one set of signal spectrum diagrams, identify the noise corresponding to the signal spectrum diagrams, and use wavelet soft threshold denoising method to remove the noise corresponding to the signal spectrum diagrams to generate a preprocessed monitoring set.

[0042] In this embodiment, the wavelet soft thresholding method is used to remove noise from the signal spectrum, which can retain key features while removing noise, thereby improving the risk identification capability of the early warning management model.

[0043] This invention provides a method for training a pre-built early warning management model based on historical monitoring information. Figure 5 A schematic diagram illustrating the implementation process of a pre-built early warning management model trained based on historical monitoring information is shown. The method includes: S401, Load the pre-built early warning management model, preset the hyperparameters, training epochs, loss function and activation function of the early warning management model. The loss function can be the cross-entropy loss function, the activation function can be the sigmoid function, the hyperparameter can be 0.002, and the training epochs are 50-150. S402, acquire historical monitoring information, use the Laplace mechanism to add noise to the historical monitoring information, and divide the noise-added historical monitoring information into training set and test set. It should be noted that the historical monitoring information can be time-series, spatial, or operational facility monitoring data. At the same time, using the Laplace mechanism to add noise to the historical monitoring information can enhance the data. The processed data can be processed with a unified timestamp. The ratio of training set to test set can be 4:1. S403 uses a training set to iteratively train the early warning management model. During iterative training, Bayesian optimization is used to adjust the model hyperparameters, calculate the output of each layer, adjust the weights and biases, minimize the prediction error, and output a converged early warning management model. In the iterative training process, Bayesian optimization is used to adjust the model hyperparameters, which can more efficiently search the hyperparameter space, find the optimal combination of hyperparameters, and improve the performance of the model. S404: Obtain the test set. The early warning management model identifies and analyzes the test set, outputs the test balance value, and identifies the influencing indicators associated with the test balance value. S405, determine whether the impact indicators associated with the test equilibrium value are consistent with the actual impact indicators. In this embodiment of the invention, determining whether the impact indicators associated with the test equilibrium value are consistent with the actual impact indicators is essentially to verify the degree of matching between the model prediction results and the actual risk. This can be comprehensively evaluated through quantitative comparison, statistical testing, visualization analysis, and other methods. Specifically, Bland-Altman analysis can be used to draw a scatter plot of the predicted values ​​and the actual values ​​to calculate the consistency boundary. If 90% of the points fall within the LoA, the consistency is considered good. Through the consistency verification of the test set, the model's generalization ability to new data can be evaluated. S406, If the impact indicators associated with the test equilibrium value are consistent with the actual impact indicators, output the converged early warning management model; If the impact index associated with the test equilibrium value is inconsistent with the actual impact index, the SGD optimizer is used to update and optimize the model hyperparameters, and the process returns to S403 to continue iterative training of the early warning management model.

[0044] In this embodiment, the early warning management model uses a convolutional neural network model as the initial model. The initial model includes an input layer, an output layer, and a convolutional network architecture. The input layer incorporates the PBFT consensus algorithm, which integrates artificial intelligence algorithms, and a weighted bidirectional feature pyramid network. This addresses the heterogeneity problem of multi-sensor data, fusing multi-source data through a consensus mechanism to eliminate device errors and improve the reliability of input data. The weighted bidirectional feature pyramid network, through bottom-up and top-down bidirectional paths combined with an attention mechanism, weights key features to enhance the model's ability to perceive "critical risk areas." The convolutional network architecture includes convolutional layers, pooling layers, and fully connected layers. The system adds a Long Short-Term Memory (LSTM) neural network after the fully connected layer, incorporating the DPoS consensus algorithm. A TCN module based on fuzzy hierarchical analysis is added between the convolutional network architecture and the output layer. The TCN module uses fuzzy hierarchical analysis combined with triangular fuzzy number theory to fuzzily couple the anomaly and risk levels of monitoring points in the preprocessed monitoring set, outputting dynamic equilibrium values ​​for monitoring control points. It captures long-range temporal dependencies through dilated convolutions, thus compensating for the shortcomings of LSTM in modeling ultra-long-term dependencies. Furthermore, the combination of fuzzy hierarchical analysis and the TCN module integrates expert experience with data-driven approaches, outputting dynamic equilibrium values ​​that better meet engineering requirements and reducing false alarm rates.

[0045] In this embodiment of the invention, the early warning management model uses a convolutional neural network model as the initial model. The initial model incorporates the PBFT consensus algorithm, which integrates artificial intelligence algorithms, and a weighted bidirectional feature pyramid network. This overcomes data conflicts caused by equipment errors and transmission delays from different sensors in water conservancy monitoring. A long short-term memory neural network is added after the fully connected layer, enabling simultaneous processing of spatial and temporal features, making it suitable for the complex characteristics of water conservancy facility monitoring data. The DPoS consensus algorithm is introduced into the long short-term memory neural network, and the combination of fuzzy hierarchical analysis and the TCN module integrates expert experience with data-driven approaches, outputting dynamic equilibrium values ​​that better meet engineering requirements. Furthermore, the integration of triangular fuzzy number theory further enhances the model's ability to handle uncertainty. This allows for more accurate description and processing of fuzzy information in monitoring data, improving the model's robustness and significantly reducing the false alarm rate.

[0046] This invention provides a method for identifying and analyzing preprocessed monitoring sets using an early warning management model. Figure 6 The diagram illustrates the implementation flow of the pre-processed monitoring set identification and analysis method of the early warning management model. The method includes: S501, obtain the preprocessed monitoring set. The input layer pre-encodes the facility monitoring data in the preprocessed monitoring set based on the PBFT consensus algorithm. PBFT is highly robust to malicious attacks or random noise, avoiding model misjudgment caused by single-source data pollution, which facilitates water conservancy management. S502, the weighted bidirectional feature pyramid network, fuses pre-encoded facility monitoring data in a bottom-up manner and outputs the data fusion features corresponding to the monitoring control points. It should be noted that during data fusion, feature fusion at different levels can be achieved by combining shallow local anomalies and deep global trends. This allows the model to simultaneously capture multi-scale risks such as local sudden leakage and global water level exceeding limits, avoiding missed detections caused by focusing on only a single scale. S503 assigns the comprehensive weight of the monitoring and control points to the data fusion features of the monitoring and control points. The convolutional network architecture extracts at least one set of local anomaly features from the data fusion features and performs nonlinear activation on the local anomaly features to obtain the activated anomaly feature map. A weighted adjacency matrix is ​​then constructed based on the activated anomaly feature map. The weighted adjacency matrix constructed based on the activated anomaly feature map can intuitively show the propagation link of risk between monitoring points. S504, the fully connected layer uses the Leiden algorithm to calculate the weighted adjacency matrix to obtain the anomaly degree of the monitoring points; S505 loads the anomaly score of the monitoring point. The long short-term memory neural network predicts and assesses the risk of the monitoring control point based on the anomaly score of the monitoring point and the data fusion features, and outputs the risk score of the monitoring control point corresponding to the monitoring point. The S506 TCN module, based on fuzzy hierarchical analysis and triangular fuzzy number theory, fuzzily couples the anomaly degree and risk degree of monitoring points in the preprocessed monitoring set, and outputs the dynamic balance value of the monitoring control points. It should be noted that triangular fuzzy number theory can transform fuzzy natural language rules into calculable numerical values ​​and output continuous dynamic balance values, thereby significantly reducing the false alarm rate and false alarm rate, and supporting the safe operation and precise management of water conservancy projects.

[0047] This invention provides a method for constructing a domino effect propagation model based on anomaly monitoring points and cascading failures. Figure 7 A schematic diagram illustrating the implementation process of a method for constructing a domino effect propagation model combining cascading failures based on anomaly monitoring points is shown. The method includes: S601. Obtain at least one set of abnormal monitoring points, construct an undirected graph of risk propagation by combining the facility geographic parameters corresponding to the abnormal monitoring points, and add the dynamic balance value of the abnormal monitoring points to the undirected graph of risk propagation. In the undirected graph of risk propagation, the graph nodes represent abnormal monitoring points, and the weight of the graph connecting edges represents the risk propagation intensity. It should be noted that transforming discrete abnormal monitoring points and their spatial relationships into a graph structure can intuitively show the physical links of risk propagation and provide a clear path framework for subsequent analysis. S602 uses a Bayesian network probabilistic model to calculate the probability of risk propagation between nodes and introduces a time decay factor into the undirected graph of risk propagation. It quantifies the failure of risk propagation through a Markov logic network. The Bayesian network can accurately describe the uncertainty of risk propagation between nodes through probability distribution, thereby avoiding the black-and-white defects of traditional deterministic models and more closely reflecting the actual distribution of risks in water conservancy facilities. The introduction of the time decay factor simulates the decay efficiency of risk, while the Markov logic network quantifies failure through logical rules, enabling the model to dynamically reflect the decay or enhancement mechanism in the risk propagation process. S603, load Markov logic network and risk propagation undirected graph, fuse Markov logic network and risk propagation undirected graph to obtain domino effect propagation model; S604 extracts impact indicators of water conservancy accidents through a domino effect propagation model and extracts cascading risk impacts of water conservancy facility topology based on a cascading failure algorithm. It also provides feedback on the cascading risk impacts of water conservancy facility topology. The cascading failure algorithm can quantify the evolution process from single-point anomalies to global anomalies and visualize it in the form of heat maps, risk level maps, etc., to help managers quickly formulate emergency strategies.

[0048] In this embodiment, the graph structure of the undirected risk propagation graph can intuitively display the propagation path. Combined with the probability model to quantify uncertainty, the early warning management model can not only predict whether the risk will occur, but also explain how the risk will occur and demonstrate the risk propagation path. This provides water conservancy experts with verifiable risk evolution logic, improves the credibility of the early warning, and extracts key impact indicators and risk propagation paths through cascading effect analysis. The model can directly output decision suggestions such as a list of high-risk nodes and the optimal reinforcement sequence.

[0049] This invention provides a water conservancy digital management system that integrates artificial intelligence. Figure 8 A schematic diagram of a water conservancy digital management system integrating artificial intelligence is shown. The system includes: The graph topology module 100 is used to acquire basic information and historical monitoring information of water conservancy facilities, identify candidate control points in water conservancy facilities, assign weights to candidate control points through a combination weighting method, determine monitoring control points based on the combination weighting results of candidate control points, and generate a graph topology of water conservancy facilities. The graph topology module 100 includes: The control point candidate unit 110 is used to acquire basic information of water conservancy facilities, identify the geographical parameters of facilities in the basic information of water conservancy facilities, and generate candidate control points based on the geographical parameters of facilities and historical monitoring information for monitoring the impact indicators of structural safety, seepage safety, stress-strain safety, operating status, environmental correlation, water use efficiency, soil moisture, drainage network and bank slope stability of water conservancy facilities. Subjective weighting unit 120 uses structural safety, seepage safety, and risk correlation as indicator layers. It performs subjective weighting on candidate control points based on the analytic hierarchy process. It constructs a subjective weight matrix through pairwise comparisons and uses consistency checks to obtain the subjective weights of the candidate control points. Objective weighting unit 130 determines the coefficient of variation of the influence indicators associated with candidate control points based on the coefficient of variation method, and performs objective weighting based on the coefficient of variation of the influence indicators associated with candidate control points to obtain the objective weight of candidate control points. The comprehensive weighting unit 140 is used to load the subjective weights and objective weights of the candidate control points. Based on game theory, the comprehensive weight including subjective and objective weights is determined. When calculating the comprehensive weight, the subjective weights and objective weights of the candidate control points are linearly combined. The coefficients of the linear combination are optimized based on game theory to calculate the optimal comprehensive weight, thereby obtaining the comprehensive weight of the candidate control points and outputting the combined weighting result containing the comprehensive weight. The control point determination unit 150 is used to traverse the combined weighting results and determine whether the comprehensive weight of the candidate control point exceeds the preset weighting threshold. If the comprehensive weight of the candidate control point exceeds the preset weighting threshold, the candidate control point is set as a monitoring control point. If the comprehensive weight of the candidate control point does not exceed the preset weighting threshold, the candidate control points are merged by combining the objective weight and spatiotemporal correlation of the candidate control points, and the merged control point result is set as a monitoring control point. The graph topology construction unit 160 is used to load at least one set of monitoring and control points and identify the facility geographic parameters of the monitoring and control points, and to construct a water conservancy facility graph topology containing the monitoring and control points using a 3D modeling artifact.

[0050] The monitoring data acquisition module 200 is used to acquire facility monitoring data collected from monitoring control points in the water conservancy facility map topology in real time, preprocess the facility monitoring data, and output a preprocessed monitoring set. The monitoring and analysis module 300 trains a pre-built early warning management model based on historical monitoring information, loads a pre-processed monitoring set, identifies and analyzes the pre-processed monitoring set, outputs the dynamic balance value of the monitoring control point, and compares and analyzes the dynamic balance value of the monitoring control point with the preset risk threshold to determine whether the monitoring control point is abnormal. If the monitoring control point is determined to be an abnormal monitoring point, an abnormal monitoring point instruction is triggered. The cascade analysis module 400 is used to acquire at least one set of abnormal monitoring points, construct a domino effect propagation model based on the abnormal monitoring points and combine cascade failures, predict the cascade risk impact of the water conservancy facility map topology, and provide feedback on the cascade risk impact of the water conservancy facility map topology.

[0051] It should be noted that the graph topology module 100, monitoring data acquisition module 200, monitoring analysis module 300, and cascade analysis module 400 of the water conservancy data management system integrating artificial intelligence in this embodiment of the invention correspond to the above-mentioned water conservancy data management method integrating artificial intelligence, and will not be described again here.

[0052] In summary, this invention provides a water conservancy digital management system and method that integrates artificial intelligence. In the embodiments of this invention, when determining monitoring control points and generating a water conservancy facility map topology based on the combined weighting results of candidate control points, a comprehensive evaluation of candidate control points is conducted through a combined weighting method. By combining subjective and objective weighting and introducing game theory to optimize the comprehensive weight, the geographical parameters, historical monitoring information, and dynamic changes of various indicators of water conservancy facilities can be considered more comprehensively. This allows for more accurate selection of monitoring control points, making the final determined monitoring control points more consistent with the actual risk distribution and covering key risk areas. This significantly improves the intelligence level and scientific decision-making of water conservancy digital management.

[0053] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0054] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. A water conservancy digital management method integrating artificial intelligence, characterized in that, The method includes: Acquire basic information and historical monitoring information of water conservancy facilities, identify candidate control points in water conservancy facilities, assign weights to candidate control points using a combination weighting method, determine monitoring control points based on the combination weighting results of candidate control points, and generate a topology map of water conservancy facilities. Real-time acquisition of facility monitoring data collected from monitoring and control points in the water conservancy facility map topology; preprocessing of facility monitoring data; output of preprocessed monitoring set. A pre-built early warning management model is trained based on historical monitoring information. A pre-processed monitoring set is loaded, and the early warning management model identifies and analyzes the pre-processed monitoring set to output the dynamic balance value of the monitoring control points. The dynamic balance value of the monitoring control point is compared and analyzed with the preset risk threshold to determine whether the monitoring control point is abnormal. If the monitoring control point is determined to be an abnormal monitoring point, an abnormal monitoring point command is triggered. Obtain at least one set of abnormal monitoring points, construct a domino effect propagation model based on the abnormal monitoring points and combine cascading failures, predict the cascading risk impact of the water conservancy facility map topology, and provide feedback on the cascading risk impact of the water conservancy facility map topology.

2. The water conservancy digital management method integrating artificial intelligence as described in claim 1, characterized in that: The method for assigning combined weights to candidate control points using a combined weighting method includes: Acquire basic information on water conservancy facilities, identify the geographical parameters of the facilities in the basic information, and generate candidate control points based on the geographical parameters and historical monitoring information for monitoring the structural safety, seepage safety, stress-strain safety, operational status, environmental impact, water use efficiency, soil moisture, drainage network, and bank slope stability of water conservancy facilities. Using structural safety, seepage safety, and risk correlation as indicator layers, the candidate control points are subjectively weighted based on the analytic hierarchy process. A subjective weight matrix is ​​constructed through pairwise comparisons, and a consistency test is used to obtain the subjective weights of the candidate control points. The coefficient of variation of the influence indicators associated with candidate control points is determined based on the coefficient of variation method. Objective weights are then assigned to the candidate control points based on the coefficient of variation of the influence indicators associated with them. The subjective and objective weights of candidate control points are loaded, and the comprehensive weight including subjective and objective weights is determined based on game theory. When calculating the comprehensive weight, the subjective and objective weights of candidate control points are linearly combined, and the coefficients of the linear combination are optimized based on game theory to calculate the optimal comprehensive weight, thereby obtaining the comprehensive weight of candidate control points. The combined weighting result including the comprehensive weight is output. Iterate through the combined weighting results and determine whether the comprehensive weight of the candidate control points exceeds the preset weighting threshold; If the overall weight of a candidate control point exceeds the preset weighting threshold, the candidate control point will be set as a monitoring control point. If the overall weight of the candidate control points does not exceed the preset weighting threshold, the candidate control points are merged based on their objective weights and spatiotemporal correlation, and the merged control point result is set as the monitoring control point. Load at least one set of monitoring and control points and identify the facility geographic parameters of the monitoring and control points. Use a 3D modeling artifact to construct a topology map of water conservancy facilities containing the monitoring and control points.

3. The water conservancy digital management method integrating artificial intelligence as described in claim 2, characterized in that: The method for merging candidate control points by combining objective weights and spatiotemporal correlation includes: The objective weights of candidate control points are identified, the coefficient of variation of candidate control points is extracted by the back-inference method, and the control point information entropy, which represents the coefficient of variation of candidate control points and the comprehensive information of facility geographical parameters, is calculated based on the normal distribution. Load the control point information entropy of the candidate control point, determine whether the control point information entropy exceeds the preset spatiotemporal information threshold, and if the control point information entropy exceeds the preset spatiotemporal information threshold, set the candidate control point as the center control point. Based on the control point information entropy of the central control point, a control point spatial weight matrix and a time correlation coefficient matrix are constructed, where the spatial weight matrix is ​​a distance decay matrix and the time correlation coefficient matrix is ​​a Pearson correlation coefficient matrix. Multiply the spatial weight matrix of the control points and the temporal correlation coefficient matrix to obtain the comprehensive spatiotemporal correlation matrix. Analyze the comprehensive spatiotemporal correlation matrix and select 40% of the quantile of the comprehensive spatiotemporal correlation matrix as the mutual information threshold. Based on spatiotemporal correlation analysis, the spatiotemporal correlation mutual information between the central control point and the associated control points is calculated. It is then determined whether the spatiotemporal correlation mutual information between the central control point and the associated control points is greater than the mutual information threshold. If the spatiotemporal correlation mutual information between the central control point and the associated control points is greater than the mutual information threshold, the associated control points and the central control point are merged, and the merged control point result is set as the monitoring control point.

4. The water conservancy digital management method integrating artificial intelligence as described in claim 1, characterized in that: The method for preprocessing facility monitoring data includes: Acquire facility monitoring data, remove outliers from the facility monitoring data, and output the facility monitoring data after outlier removal. Perform spectrum analysis on the facility monitoring data after outlier processing, and adaptively decompose the facility monitoring data based on the spectrum analysis results to obtain at least one set of monitoring signal segments; The monitoring data of facilities within the monitoring signal segment is converted into the frequency domain to obtain the signal spectrum. The energy distribution of the signal spectrum is calculated, and the energy value corresponding to the noise in the signal spectrum is output. Load at least one set of signal spectrograms, identify the noise corresponding to the signal spectrograms, and use wavelet soft threshold denoising to remove the noise corresponding to the signal spectrograms to generate a preprocessed monitoring set.

5. The water conservancy digital management method integrating artificial intelligence as described in claim 4, characterized in that: The method for training a pre-built early warning management model based on historical monitoring information includes: Load a pre-built early warning management model, and preset the hyperparameters, training rounds, loss function, and activation function of the early warning management model; Historical monitoring information is acquired, and the Laplace mechanism is used to add noise to the historical monitoring information. The noise-added historical monitoring information is then divided into a training set and a test set. The early warning management model is iteratively trained using a training set. During iterative training, Bayesian optimization is used to adjust the model hyperparameters, and the output of each layer is calculated. The weights and biases are adjusted to minimize the prediction error and output a converged early warning management model. The test set is acquired, and the early warning management model identifies and analyzes the test set, outputs the test balance value, identifies the impact indicators associated with the test balance value, and determines whether the impact indicators associated with the test balance value are consistent with the actual impact indicators. If the impact indicators associated with the test balance value are consistent with the actual impact indicators, the converged early warning management model is output. If the impact indicators associated with the test equilibrium value are inconsistent with the actual impact indicators, the SGD optimizer is used to update and optimize the model hyperparameters, and the early warning management model is iteratively trained.

6. The water conservancy digital management method integrating artificial intelligence as described in claim 5, characterized in that: The early warning management model uses a convolutional neural network model as the initial model. The initial model includes an input layer, an output layer, and a convolutional network architecture. The input layer introduces the PBFT consensus algorithm, which integrates artificial intelligence algorithms, and a weighted bidirectional feature pyramid network. The convolutional network architecture includes convolutional layers, pooling layers, and fully connected layers. A long short-term memory neural network is added after the fully connected layers. The DPoS consensus algorithm is introduced into the long short-term memory neural network. A TCN module based on fuzzy hierarchical analysis is added between the convolutional network architecture and the output layer. The TCN module uses fuzzy hierarchical analysis combined with triangular fuzzy number theory to fuzzily couple the anomaly degree and risk degree of the monitoring points in the preprocessed monitoring set, and outputs the dynamic balance value of the monitoring control points.

7. The water conservancy digital management method integrating artificial intelligence as described in claim 6, characterized in that: The early warning management model includes a pre-processing monitoring set identification and analysis method, comprising: The preprocessed monitoring set is obtained, and the input layer precodes the facility monitoring data in the preprocessed monitoring set based on the PBFT consensus algorithm. The weighted bidirectional feature pyramid network fuses pre-encoded facility monitoring data in a bottom-up manner and outputs the data fusion features corresponding to the monitoring control points. The comprehensive weight of the monitoring and control points is assigned to the data fusion feature of the monitoring and control points. The convolutional network architecture extracts at least one set of local anomaly features from the data fusion feature, and performs nonlinear activation on the local anomaly features to obtain the activated anomaly feature map. A weighted adjacency matrix is ​​then constructed based on the activated anomaly feature map. The fully connected layer uses the Leiden algorithm to calculate the weighted adjacency matrix to obtain the anomaly degree of the monitoring points; The monitoring point anomaly score is loaded, and the long short-term memory neural network predicts and assesses the risk of the monitoring control point based on the monitoring point anomaly score and data fusion features, and outputs the monitoring point risk score corresponding to the monitoring control point. The TCN module uses fuzzy hierarchical analysis combined with triangular fuzzy number theory to fuzzily couple the anomaly degree and risk degree of monitoring points in the preprocessed monitoring set, and outputs the dynamic balance value of the monitoring control points.

8. The water conservancy digital management method integrating artificial intelligence as described in claim 7, characterized in that: The method for constructing a domino effect propagation model based on anomaly monitoring points and incorporating cascading failures includes: Obtain at least one set of abnormal monitoring points, construct an undirected graph of risk propagation by combining the facility geographic parameters corresponding to the abnormal monitoring points, and add the dynamic balance value of the abnormal monitoring points to the undirected graph of risk propagation. In the undirected graph of risk propagation, the graph nodes represent abnormal monitoring points, and the weight of the graph connecting edges represents the risk propagation intensity. A Bayesian network probability model is used to calculate the probability of risk propagation between nodes, and a time decay factor is introduced into the undirected graph of risk propagation. The failure of risk propagation is quantified by Markov logic network. By loading a Markov logic network and an undirected graph of risk propagation, and fusing the Markov logic network and the undirected graph of risk propagation, a domino effect propagation model is obtained. The impact indicators of water conservancy accidents are extracted using a domino effect propagation model, and the cascading risk impact of the water conservancy facility topology is extracted based on a cascading failure algorithm, thus providing feedback on the cascading risk impact of the water conservancy facility topology.

9. A water conservancy digital management system integrating artificial intelligence, used to implement the water conservancy digital management method integrating artificial intelligence as described in any one of claims 1-8, characterized in that: The water conservancy digital management system that integrates artificial intelligence includes: The graph topology module is used to acquire basic information and historical monitoring information of water conservancy facilities, identify candidate control points in water conservancy facilities, assign weights to candidate control points through a combination weighting method, determine monitoring control points based on the combination weighting results of candidate control points, and generate a graph topology of water conservancy facilities. The monitoring data acquisition module is used to acquire facility monitoring data collected from monitoring control points in the water conservancy facility map topology in real time, preprocess the facility monitoring data, and output a preprocessed monitoring set. The monitoring and analysis module trains a pre-built early warning management model based on historical monitoring information, loads a pre-processed monitoring set, identifies and analyzes the pre-processed monitoring set, outputs the dynamic balance value of the monitoring control point, and compares and analyzes the dynamic balance value of the monitoring control point with the preset risk threshold to determine whether the monitoring control point is abnormal. If the monitoring control point is determined to be an abnormal monitoring point, an abnormal monitoring point command is triggered. The cascade analysis module is used to acquire at least one set of abnormal monitoring points, construct a domino effect propagation model based on the abnormal monitoring points and cascade failures, predict the cascade risk impact of the water conservancy facility map topology, and provide feedback on the cascade risk impact of the water conservancy facility map topology.

10. The water conservancy digital management system integrating artificial intelligence as described in claim 9, characterized in that: The graph topology module includes: The control point candidate unit is used to obtain basic information of water conservancy facilities, identify the geographical parameters of the facilities in the basic information of water conservancy facilities, and generate candidate control points based on the geographical parameters of the facilities and historical monitoring information for monitoring the impact indicators of structural safety, seepage safety, stress-strain safety, operating status, environmental correlation, water use efficiency, soil moisture, drainage network and bank slope stability of water conservancy facilities. The subjective weighting unit uses structural safety, seepage safety, and risk correlation as indicator layers. It subjectively weights candidate control points based on the analytic hierarchy process, constructs a subjective weight matrix through pairwise comparisons, and obtains the subjective weights of candidate control points using a consistency test. The objective weighting unit determines the coefficient of variation of the influence indicators associated with the candidate control points based on the coefficient of variation method, and performs objective weighting based on the coefficient of variation of the influence indicators associated with the candidate control points to obtain the objective weight of the candidate control points. The comprehensive weighting unit loads the subjective and objective weights of the candidate control points, determines the comprehensive weight including subjective and objective weights based on game theory, and calculates the comprehensive weight including subjective and objective weights by linearly combining the subjective and objective weights of the candidate control points during the comprehensive weight calculation. The coefficients of the linear combination are optimized based on game theory to calculate the optimal comprehensive weight, obtain the comprehensive weight of the candidate control points, and output the combined weighting result including the comprehensive weight. The control point determination unit is used to traverse the combined weighting results and determine whether the comprehensive weight of the candidate control point exceeds the preset weighting threshold. If the comprehensive weight of the candidate control point exceeds the preset weighting threshold, the candidate control point is set as a monitoring control point. If the comprehensive weight of the candidate control point does not exceed the preset weighting threshold, the candidate control points are merged by combining the objective weight and spatiotemporal correlation of the candidate control points, and the merged control point result is set as a monitoring control point. The graph topology construction unit is used to load at least one set of monitoring and control points and identify the facility geographic parameters of the monitoring and control points. It uses a 3D modeling artifact to construct a water conservancy facility graph topology containing the monitoring and control points.

Images