CN-121981557-A - Power transmission and transformation foundation flood vulnerability assessment method and system based on optimized neural network

CN121981557ACN 121981557 ACN121981557 ACN 121981557ACN-121981557-A

Abstract

The invention discloses a power transmission and transformation foundation flood vulnerability assessment method and system based on an optimized neural network, and relates to the technical field of power facility disaster assessment, wherein the method comprises the steps of obtaining structural parameters and flood boundary information, and forming working condition data by combining flushing dynamic time sequence characteristics; the method comprises the steps of mapping working condition data to a two-dimensional phase space, identifying a working condition set to be supplemented with insufficient coverage of a sample, constructing a physical constraint item based on a flow-solid-seepage coupling equation, generating counternetwork supplementing working condition data to form a training data set by utilizing physical information, constructing a vulnerability assessment network, performing iterative training, constructing a negative curvature direction based on the maximum Lyapunov index monitoring result distribution when the distribution tends to be concentrated, updating network parameters according to a second-order gradient of a loss function, inputting real-time flood load into the updated network, and outputting a instability risk level after forward checking through the coupling equation. The invention solves the problems of insufficient coverage of samples and unsound combination of physical mechanism and data driving, and realizes high-precision risk assessment.

Inventors

YU TAO
LI WENWU
LIU YANG
CHEN SHENGPENG
XU XUEFENG
ZHANG XUECHAO
SI HUI
DING SHAOJUN
LUO YAFEI
XIA YANG
CHEN HOUSHENG
CHEN YINGYING

Assignees

安徽鸿霁科技有限公司
国网安徽省电力有限公司池州供电公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. The power transmission and transformation foundation flood vulnerability assessment method based on the optimized neural network is characterized by comprising the following steps of: Acquiring structural parameters and flood boundary information of a power transmission and transformation foundation, and forming working condition data by combining dynamic time sequence characteristics of scour evolution; mapping the working condition data to a two-dimensional phase space, and identifying a working condition set to be supplemented, which is insufficient in sample coverage; constructing a physical constraint item based on a fluid-solid-seepage coupling equation, and generating counternetwork supplementary working condition data by utilizing physical information to form a training data set; Constructing a vulnerability assessment network and performing iterative training, and constructing a negative curvature direction and updating network parameters according to a second-order gradient of a loss function when distribution tends to be concentrated based on the maximum Lyapunov exponent monitoring result distribution; And inputting the real-time flood load into the updated network, and outputting the instability risk level after forward checking through the coupling equation.
2. The method for evaluating the vulnerability of the flooding of the power transmission and transformation foundation based on the optimized neural network according to claim 1, wherein the obtaining the structural parameters and the flooding boundary information of the power transmission and transformation foundation comprises: The method comprises the steps of obtaining design burial depth, foundation type and soil parameters of a power transmission and transformation foundation as structural parameters, and obtaining water level, flow speed or flow speed gradient of a surrounding water area as flood boundary information; Converting flood boundary information into local scour driving force acting on soil mass around the foundation, and converting the structural parameters into scour resistance of the foundation; calculating to obtain a scalar characteristic representing the relative scouring risk of the foundation according to the ratio of the local scouring driving force to the scouring resistance; And correlating the scalar features with the corresponding scouring depth observation values to form working condition data.
3. The method for evaluating vulnerability of flood on power transmission and transformation foundation based on optimized neural network according to claim 2, wherein the correlating scalar features with corresponding observed values of flushing depth to form working condition data comprises: Acquiring local scour driving force data under a continuous time sequence in a flood process, and constructing a time sequence of the evolution of the scour depth along with time; performing empirical mode decomposition on the time sequence, and extracting intrinsic mode function components representing progressive destruction characteristics of soil; and the instantaneous frequency and the instantaneous amplitude of the eigenmode function component are used as dynamic time sequence characteristics and are spliced with the scalar characteristics to form working condition data.
4. The method for evaluating vulnerability of power transmission and transformation foundation flood based on optimized neural network according to claim 3, wherein the mapping the working condition data to the two-dimensional phase space, and identifying the working condition set to be supplemented with insufficient coverage of the sample specifically comprises: Mapping the working condition data to a two-dimensional phase space formed by flood action intensity characteristics and soil critical shear stress; determining a scouring initiation boundary in the two-dimensional phase space, and determining a statistical region according to the scouring initiation boundary; gridding the statistical region, and calculating the sample density of each grid unit; and determining a working condition interval corresponding to the grid unit with the sample density lower than the preset threshold as a working condition set to be supplemented.
5. The method for evaluating the flooding vulnerability of the power transmission and transformation foundation based on the optimized neural network according to claim 4, wherein the physical constraint term is constructed based on a flow-solid-seepage coupling equation, specifically: Constructing a flow-solid coupling control equation for describing interaction of flood water flow and soil around a foundation, and introducing damage variables for representing damage of a mesostructure in the soil into the equation; and constructing an unstable seepage field equation reflecting the flood level drop, and coupling seepage volume force as an additional source term into the flow-solid coupling control equation to form a flow-solid-seepage coupling equation.
6. The method for evaluating vulnerability of power transmission and transformation foundation flood based on optimized neural network according to claim 5, wherein the generating the network supplementing condition countering data by using the physical information specifically comprises: constructing and generating an countermeasure network, and extracting a solution condition of the fluid-solid-seepage coupling equation at the boundary of the working condition set to be supplemented; Introducing the flow-solid-osmotic coupling equation and the solution condition as physical constraint terms to generate a loss function of the countermeasure network; training a generating countermeasure network to ensure that the generated data simultaneously meets the physical constraint items and is consistent with the original working condition data distribution; And generating newly-increased working condition data covering the insufficient sample interval in batches by utilizing the generated countermeasure network after training convergence, and combining the newly-increased working condition data with the original working condition data to form a training data set.
7. The power transmission and transformation foundation flood vulnerability assessment method based on optimized neural network according to claim 6, wherein the monitoring result distribution based on the maximum lyapunov exponent comprises: Constructing an initial vulnerability assessment network, and training the assessment network by using a training data set; in each iteration, performing dimension reduction processing on the output of the appointed hidden layer of the evaluation network to obtain a state vector in a low-dimension embedded space; Calculating a maximum lyapunov exponent in the low-dimensional embedding space based on the time series of state vectors; when the Lyapunov exponent is negative from positive rotation and the absolute value exceeds a preset threshold, the evaluation result distribution tends to be concentrated and is about to enter a local stable state.
8. The method for evaluating the vulnerability of the power transmission and transformation foundation flood based on the optimized neural network according to claim 7, wherein the constructing the negative curvature direction and updating the network parameters according to the second-order gradient of the loss function specifically comprises: When the local stable state is about to be entered, freezing and evaluating partial network layer parameters of the current batch of the network; Calculating second-order gradient information of the loss function relative to the unfrozen layer parameters, and constructing a negative curvature direction according to the second-order gradient information; Updating the unfrozen layer parameters along the negative curvature direction to enable the evaluation network to escape from the current local area; After escape is completed, all network layer parameters are thawed, and conventional one-step updating is recovered.
9. The method for evaluating the flooding vulnerability of the power transmission and transformation foundation based on the optimized neural network according to claim 8, wherein the outputting the destabilizing risk level after forward checking through the coupling equation specifically comprises: inputting the real-time flood load into an updated evaluation network, and obtaining a basic load degradation basic predicted value through forward propagation calculation; inputting the basic pre-estimated value into the fluid-solid-seepage coupling equation to perform forward solving to obtain a theoretical failure mode corresponding to the current load; Carrying out pattern matching on the statistical failure mode output by the evaluation network and the theoretical failure mode, and calculating the matching degree; If the matching degree is lower than a preset threshold, random field disturbance is carried out on the input load, the matching frequency of the output of each disturbance sample and the theoretical failure mode is counted, and the mode with the highest matching frequency is used as the final output instability risk level.
10. A system using the power transmission and transformation foundation flooding vulnerability assessment method based on optimized neural network as claimed in any one of claims 1-9, comprising: the data acquisition unit is used for acquiring structural parameters of the power transmission and transformation foundation and flood boundary information, and forming working condition data by combining dynamic time sequence characteristics of scouring evolution; the working condition identification unit is used for mapping the working condition data to a two-dimensional phase space and identifying a working condition set to be supplemented, which is insufficient in sample coverage; The data enhancement unit is used for constructing a physical constraint item based on a flow-solid-seepage coupling equation, generating the counternetwork supplementary working condition data by utilizing physical information, and forming a training data set; The network training unit is used for constructing a vulnerability evaluation network and performing iterative training, and based on the distribution of the maximum Lyapunov exponent monitoring result, when the distribution tends to be concentrated, constructing a negative curvature direction according to the second-order gradient of the loss function and updating network parameters; and the evaluation output unit is used for inputting the real-time flood load into the updated network, and outputting the instability risk level after forward checking through the coupling equation.

Description

Power transmission and transformation foundation flood vulnerability assessment method and system based on optimized neural network Technical Field The invention relates to the technical field of power facility disaster evaluation, in particular to a power transmission and transformation foundation flood vulnerability evaluation method and system based on an optimized neural network. Background The power transmission and transformation project is a backbone network of the power system, and the basic stability of the power transmission and transformation project is directly related to the safe operation of the power grid. The vulnerability of the power transmission and transformation foundation under the action of flood is accurately estimated, and the method has important significance for disaster prevention and reduction. At present, flood vulnerability assessment methods for infrastructure are mainly divided into two types, namely a numerical simulation method based on a physical model and an intelligent assessment method based on data driving. The intelligent evaluation method based on data driving establishes the mapping relation between the disaster load and the structural response through the machine learning model, and has the advantage of high calculation efficiency. The bridge vulnerability assessment method based on the debris flow dynamic process, for example, is disclosed by the invention with the publication number of CN116910846B, and comprises the steps of obtaining a quantitative relation between the debris flow intensity and pier top displacement through training and assessment of an artificial neural network model, obtaining a pier top displacement probability distribution function through statistical analysis of pier top displacement of a plurality of samples obtained by utilizing the artificial neural network model, and obtaining failure probability of a bridge pier by combining vulnerability index grading standards. The method can timely master the probability that the bridge is damaged to different degrees under the impact of the mud-rock flows with different intensities, so as to timely perform disaster prevention and fighting work. In the above disclosed technical solution, at least the following technical problems exist: The existing method is mainly aimed at the dynamic response of rigid structures such as bridges and the like under the transient impact, and the coupling effect of soil body seepage field evolution and mesoscopic structure damage of a power transmission and transformation foundation under the long-term flood effect cannot be fully considered, so that the deviation between an evaluation result and an actual instability mode is caused. The conventional gradient descent algorithm is mostly adopted in the conventional neural network training process, so that the conventional neural network training process is easy to fall into local optimum, and a dynamic monitoring mechanism for the distribution convergence of network output results is lacked, so that the generalization capability of a model is limited. The present invention proposes a solution to the above-mentioned problems. Disclosure of Invention In order to overcome the defects of the prior art, the embodiment of the invention provides a power transmission and transformation foundation flood vulnerability assessment method and system based on an optimized neural network, which are used for solving the problems of insufficient sample coverage, and unsound combination of physical mechanism and data driving in the prior art by constructing a flow-solid-seepage coupling equation as physical constraint, utilizing generation of counternetwork supplement sparse working condition data and combining maximum Lyapunov index monitoring with negative curvature direction optimization network training. In order to achieve the above purpose, the present invention provides the following technical solutions: A power transmission and transformation foundation flood vulnerability assessment method based on an optimized neural network comprises the following steps of obtaining structural parameters and flood boundary information of the power transmission and transformation foundation, forming working condition data by combining dynamic time sequence characteristics of scouring evolution, mapping the working condition data to a two-dimensional phase space, identifying a working condition set to be supplemented with insufficient sample coverage, constructing a physical constraint item based on a fluid-solid-seepage coupling equation, generating counternetwork supplementing working condition data by using the physical information to form a training data set, constructing a vulnerability assessment network, performing iterative training, constructing a negative curvature direction and updating network parameters according to a second-order gradient of a loss function when distribution tends to be concentrated, inputting real-time flood lo