CN-122022504-A - Deep learning-based intelligent monitoring system and method for bank collapse risk in reservoir area

CN122022504ACN 122022504 ACN122022504 ACN 122022504ACN-122022504-A

Abstract

The invention provides a deep learning-based intelligent monitoring system and method for bank collapse risk in a reservoir area, which relate to the technical field of bank area monitoring and comprise the steps of acquiring and preprocessing all monitoring point data of a reservoir area monitoring network and constructing a monitoring point total set; the method comprises the steps of screening key base points from the information, quantifying inherent risk characterization quantities of the key base points, establishing a dynamic association network of common monitoring points and the key base points to obtain final clustering association degree of the common monitoring points relative to the key base points, calculating radiation risk values of the key base points relative to the associated common monitoring points by combining the inherent risk characterization quantities and the final clustering association degree, fusing the radiation risk values of all the key base points to obtain comprehensive risk values of the common monitoring points, constructing a directed weighted graph by taking the key base points as source points, dividing risk communities by adopting a graph neural network community discovery algorithm, and assigning and outputting unified risk grades to the monitoring points based on statistical distribution of the comprehensive risk values of the communities and the highest risk grades of the contained key base points.

Inventors

Zeng Jinxin
ZENG LINYAO
WANG YINGCONG
LUO CHUANG

Assignees

浙江中易慧能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (8)

1. The intelligent monitoring method for the bank collapse risk of the reservoir area based on deep learning is characterized by comprising the following steps of: Acquiring shore area monitoring data and shore area related data of all monitoring points in a reservoir shore monitoring network, preprocessing the shore area monitoring data, and constructing a monitoring point total set; Screening a plurality of key base points with high risk or extremely high risk grades from a total collection of monitoring points according to preset screening rules based on shore area monitoring data, and quantifying inherent risk characterization quantity of each key base point; Based on each common monitoring point except the key base points in the monitoring point total set, a dynamic association network between the common monitoring points and the key base points is established, the direct association degree is calculated, an iterative application conduction attenuation mechanism is used for determining the indirect association degree, and a chain association degree product operation is performed based on a risk conduction path in a backtracking way, so that the final clustering association degree of each common monitoring point relative to each key base point is obtained; Aiming at each key base point, introducing a correction function, a background risk attribute and shore area related data by combining the inherent risk characterization quantity of the key base point with the final clustering association degree of the corresponding common monitoring point, and calculating the radiation risk value of each key base point for independently radiating to the associated common monitoring point; Aiming at any common monitoring point, fusing all radiation risk values corresponding to the common monitoring points by using all key base points to obtain a comprehensive risk value corresponding to the common monitoring point; constructing a directed weighted graph according to the final cluster association degree and the risk conduction path by taking the key base points as source points, and dividing a risk community by adopting a community discovery algorithm based on a graph neural network; And on the basis of the statistical distribution of the comprehensive risk values in each community and the highest risk level of the key base points, assigning a unified risk level for each monitoring point and outputting the unified risk level.
2. The deep learning based pool bank collapse risk intelligent monitoring method of claim 1, wherein the risk level includes extremely high risk, medium risk and low risk; Based on the shore area monitoring data, screening a plurality of key base points with high risk or extremely high risk grades from a total collection of monitoring points according to a preset screening rule, and quantifying the inherent risk characterization quantity of each key base point comprises the following steps: Screening a plurality of candidate base points from the monitoring point total set based on shore area monitoring data and a preset screening rule to form a preliminary candidate base point set; based on each candidate base point in the preliminary candidate base point set, calculating an initial inherent risk characterization quantity corresponding to the candidate base point through a pre-trained risk assessment model; The candidate base points are taken as centers, adjacent monitoring points are found in a preset initial association radius, a local influence domain of the candidate base points is formed, simulation risk values of all adjacent monitoring points in the local influence domain are calculated based on a preset risk radiation model and initial inherent risk characterization quantity of the candidate base points, and the risk grade of the candidate base points is judged based on the distribution condition of the simulation risk values in the local influence domain; and taking the candidate base points with the risk level of high risk or extremely high risk as key base points, and calculating the inherent risk characterization quantity of each key base point.
3. The deep learning-based intelligent monitoring method for bank collapse risk in a reservoir area according to claim 2, wherein the steps of establishing a dynamic association network between the common monitoring points and the key base points based on each common monitoring point except the key base points in the total set of monitoring points, calculating the direct association degree, and determining the indirect association degree by iteratively applying a conduction attenuation mechanism include: Based on the key base point set and the common monitoring point set, constructing a dynamic association network topology taking the key base points as cores and the common monitoring points as association nodes; Based on the constructed dynamic association network, calculating the direct association degree between each key base point and each common monitoring point, and generating an initial association matrix; Based on a strong association threshold, a common monitoring point with the direct association degree higher than the strong association threshold is attributed to a first layer of association point corresponding to a key base point, and meanwhile, a point which has the highest direct association degree and has the inherent risk characterization quantity exceeding a secondary source threshold is screened out from the first layer of association points and added into a candidate secondary radiation source set; based on preset promotion conditions, common monitoring points meeting the promotion conditions are screened from the candidate secondary radiation source set to serve as formal secondary radiation sources; the key base point set and the formal secondary radiation source are used as the radiation source set of the current round, and the indirect association degree between each point in the radiation source set and each point in the set to be fixed point is calculated based on a conduction attenuation mechanism; In each round of iteration, attributing the calculated point to be fixed with the maximum indirect association degree higher than the weak association threshold value to a radiation source corresponding to the maximum indirect association degree, and recording the affiliated level and the risk conduction path of the radiation source; And stopping iteration when the set to be fixed point is not changed or reaches the preset maximum iteration number.
4. The deep learning-based reservoir bank collapse risk intelligent monitoring method according to claim 3, wherein the step of performing a chain relevance product operation based on the risk conduction path backtracking to obtain a final cluster relevance of each common monitoring point relative to each key base point comprises the steps of: and for each common monitoring point which belongs to, carrying out chain product calculation on each section of direct association or indirect association on the path according to the conduction attenuation mechanism through tracing back the risk conduction path to the key base point, so as to obtain the final clustering association of the common monitoring point relative to the key base point.
5. The intelligent monitoring method for bank collapse risk in deep learning based on claim 4, wherein calculating the radiation risk value of each key base point radiating to its associated common monitoring point comprises: based on the inherent risk characterization quantity corresponding to the key base point and the final clustering association degree of the common monitoring point relative to the key base point, carrying out correction processing on the final clustering association degree based on a preset correction function to obtain a correction association degree; calculating to obtain a preliminary radiation risk value based on the corrected association degree and the corresponding inherent risk characterization quantity; And correcting the preliminary radiation risk value based on the background risk attribute of the common monitoring point and the multidimensional correction coefficient corresponding to the shore area related data to obtain a final radiation risk value.
6. The deep learning-based reservoir bank collapse risk intelligent monitoring method according to claim 5, wherein the fusing the radiation risk values corresponding to all key base points to the common monitoring points for any common monitoring point to obtain the comprehensive risk value corresponding to the common monitoring point comprises: Acquiring risk conduction direction information between each key base point and each corresponding common monitoring point; Determining corresponding risk conduction direction vectors for each key base point generating a radiation risk value for the common monitoring point based on any common monitoring point by combining the risk conduction direction information; calculating vector sums of risk conduction direction vectors of all key base points generating radiation risk values for the common monitoring points to obtain a combined vector; And calculating the final comprehensive risk value of the common monitoring point based on the obtained resultant vector.
7. The deep learning-based intelligent monitoring method for bank collapse risk in a reservoir according to claim 6, wherein assigning a unified risk level to each monitoring point and outputting the unified risk level comprises: The method comprises the steps of using key base points as initial source nodes, using common monitoring points with determined final cluster association degrees as subsequent nodes, and constructing a directed weighted graph based on hierarchical attribution relations and risk conduction paths, wherein the direction of edges in the graph is a risk conduction direction, and the weight is the corresponding final cluster association degree; Based on the directed weighted graph, identifying node subsets with tight internal connection and sparse external connection by using a community discovery algorithm based on a graph neural network, wherein each subset forms a risk community; calculating the statistical characteristics of the comprehensive risk values of all nodes in the risk communities aiming at each identified risk community; and assigning a dominant risk level to the risk community according to the statistical characteristics and the highest risk level of the key base points existing in the risk community, wherein all nodes in the community inherit the dominant risk level.
8. Deep learning-based intelligent monitoring system for bank collapse risk of a reservoir, which realizes the method for intelligent monitoring for bank collapse risk of a reservoir based on deep learning as set forth in any one of claims 1 to 7, and is characterized by comprising the following steps: The acquisition module is used for acquiring the shore area monitoring data and the shore area related data of all monitoring points in the reservoir shore monitoring network, preprocessing the shore area monitoring data and constructing a monitoring point total set; The identification module is used for screening a plurality of key base points with high risk or extremely high risk grades from a total collection of monitoring points according to preset screening rules based on the shore area monitoring data, and quantifying inherent risk characterization quantity of each key base point; The analysis module is used for establishing a dynamic association network between the common monitoring points and the key base points based on each common monitoring point except the key base points in the monitoring point total set, calculating direct association, determining indirect association by iteratively applying a conduction attenuation mechanism, and executing chain association product operation based on a risk conduction path back trace so as to obtain final clustering association of each common monitoring point relative to each key base point; The risk radiation calculation module is used for introducing a correction function, background risk attributes and shore area related data aiming at each key base point and combining the inherent risk characterization quantity of the key base point with the final clustering association degree of the corresponding common monitoring point to calculate the radiation risk value of each key base point for independently radiating to the associated common monitoring point; the comprehensive risk assessment module is used for fusing radiation risk values corresponding to all key base points to the common monitoring points aiming at any common monitoring point to obtain a comprehensive risk value corresponding to the common monitoring point; the risk determining module takes key base points as source points, constructs a directed weighted graph according to the final cluster association degree and the risk conduction path, adopts a community discovery algorithm based on a graph neural network to divide risk communities, and finally assigns unified risk grades for all monitoring points and outputs the unified risk grades based on the statistical distribution of comprehensive risk values in all communities and the highest risk grade of the key base points.

Description

Deep learning-based intelligent monitoring system and method for bank collapse risk in reservoir area Technical Field The invention relates to the technical field of shore area monitoring, in particular to a reservoir area landslide risk intelligent monitoring system and method based on deep learning. Background The reservoir area is used as a core component of hydraulic engineering, and the bank slope stability of the reservoir area directly influences the reservoir operation, the peripheral safety and the ecological sustainability. The current reservoir bank collapse risk monitoring technology has obvious short plates, is difficult to adapt to intelligent monitoring requirements under the fusion of digital economy and high-end equipment manufacturing industry, and has the following specific problems: The existing monitoring method relies on manual intervention in multiple links, has complex overall flow and low efficiency, cannot realize real-time dynamic monitoring and accurate risk level output of the landslide risk, and cannot meet the core requirements of intelligent water conservancy management. Meanwhile, all monitoring points are generally regarded as independent evaluation units, a scientific monitoring point dynamic association network is not constructed, and risk conduction characteristics are not effectively described, so that the radiation influence of key risk base points on surrounding monitoring areas cannot be accurately evaluated, and the comprehensiveness and accuracy of risk evaluation are further affected. In summary, the existing monitoring technology is difficult to effectively classify the risk of bank collapse in the reservoir area. Therefore, the intelligent monitoring system and the intelligent monitoring method for the bank collapse risk of the reservoir area based on deep learning can realize the accurate classification of the bank collapse risk of the bank area, and greatly improve the accurate prevention and control effect of the bank collapse risk of the bank area. Disclosure of Invention In order to solve the technical problems, the invention aims to provide a deep learning-based intelligent monitoring system and method for bank collapse risk of a reservoir, which can realize the accurate classification of bank collapse risk of a bank and greatly improve the accurate prevention and control effect of bank collapse risk of the bank. In order to achieve the purpose, the invention provides the technical scheme that the intelligent monitoring method for the bank collapse risk of the reservoir area based on deep learning comprises the following steps: Acquiring shore area monitoring data and shore area related data of all monitoring points in a reservoir shore monitoring network, preprocessing the shore area monitoring data, and constructing a monitoring point total set; Screening a plurality of key base points with high risk or extremely high risk grades from a monitoring point total set according to preset screening rules based on the shore area monitoring data and shore area related data, and quantifying inherent risk characterization quantity of each key base point; Based on each common monitoring point except the key base points in the monitoring point total set, a dynamic association network between the common monitoring points and the key base points is established, the direct association degree is calculated, an iterative application conduction attenuation mechanism is used for determining the indirect association degree, and a chain association degree product operation is performed based on a risk conduction path in a backtracking way, so that the final clustering association degree of each common monitoring point relative to each key base point is obtained; Aiming at each key base point, introducing a correction function, a background risk attribute and shore area related data by combining the inherent risk characterization quantity of the key base point with the final clustering association degree of the corresponding common monitoring point, and calculating the radiation risk value of each key base point for independently radiating to the associated common monitoring point; Aiming at any common monitoring point, fusing all radiation risk values corresponding to the common monitoring points by using all key base points to obtain a comprehensive risk value corresponding to the common monitoring point; And finally, assigning and outputting unified risk grades to all monitoring points based on the statistical distribution of comprehensive risk values in all communities and the highest risk grade of the contained key base points. Preferably, the risk level includes extremely high risk, medium risk, and low risk; Based on the shore area monitoring data, screening a plurality of key base points with high risk or extremely high risk grades from a total collection of monitoring points according to a preset screening rule, and quantifying the inherent risk characterization quan