CN-121503931-B - Insulator pollution risk assessment and prediction method, system, equipment and medium

CN121503931BCN 121503931 BCN121503931 BCN 121503931BCN-121503931-B

Abstract

The application discloses a method, a system, equipment and a medium for evaluating and predicting pollution risk of an insulator, belonging to the technical field of state evaluation and intelligent operation and maintenance of an electric power system; the method comprises the steps of constructing continuous pollution risk indexes for each target node, constructing causal prior information for describing the action relation according to the action relation among pollution source emission, meteorological conditions, factory layout and insulator positions, and training a pre-established space-time prediction model by utilizing a multi-dimensional time sequence feature vector and corresponding continuous pollution risk indexes to obtain an insulator pollution risk prediction model. The application can more fully utilize multi-source data related to the risk of the outdoor insulator to construct pollution risk indexes with definite physical significance, introduces causal prior constraint driven by a mechanism and improves the precision and reliability of the pollution risk prediction of the insulator.

Inventors

HUA ZEXI
QI ZHIPENG
SU ZHONGJI
QIAO HUI
RUAN BO

Assignees

西南交通大学

Dates

Publication Date: 20260512
Application Date: 20260114

Claims (8)

1. The method for evaluating and predicting the pollution risk of the insulator is characterized by comprising the following steps of: Acquiring multi-source data related to pollution risk of an outdoor insulator, preprocessing the multi-source data, and constructing a time-space aligned multi-dimensional time sequence feature vector, wherein the multi-source data comprises insulator state data, pollution source emission data and meteorological environment data; Constructing a continuous pollution risk index for each target node based on the multi-dimensional time sequence feature vector, and taking the continuous pollution risk index as a pollution risk intensity label of the corresponding target node at each moment; constructing causal prior information for describing the action relationship according to the action relationship among pollution source emission, meteorological conditions, factory layout and insulator positions; Training a pre-established space-time prediction model by utilizing the multi-dimensional time sequence feature vector and the corresponding continuous pollution risk index to obtain an insulator pollution risk prediction model, wherein the space-time prediction model comprises a node feature coding part, a space-dependent modeling part and a time-dependent modeling part, the space-dependent modeling part models the space dependence among nodes under the causal priori information constraint, and the time-dependent modeling part carries out time modeling on a space feature sequence of a target node in a preset time window to output a continuous pollution risk index predicted value at a predicted moment; Collecting new multi-source data related to the pollution risk of the outdoor insulator in real time, preprocessing, updating the multi-dimensional time sequence feature vector, inputting the updated multi-dimensional time sequence feature vector into the pollution risk prediction model of the insulator, and outputting a pollution risk prediction result of the insulator; the construction process of the causal priori information comprises the following steps: The method comprises the steps of constructing node types and node sets, namely abstracting related entities in a scene into nodes of different types, forming the node sets, and associating the node types of each node in the node sets, wherein the node types comprise a transformer station/insulator node type, a pollution source node type and a meteorological environment node type; the relation type is constructed by defining a plurality of relation types among nodes to form a relation set according to a physical mechanism and engineering experience, wherein the relation set comprises a pollutant transportation relation, an environment similarity relation and a factory topology or electrical connection relation; constructing a structure mask matrix, namely constructing the structure mask matrix for each relation type in the relation set, wherein elements in the structure mask matrix are used for representing whether directed edges pointing from one node to another node are allowed to exist under the corresponding relation type, if so, the elements are set to be 1, and if not, the elements are set to be 0; The construction of side weight priori information, namely defining side weight priors according to the distance between nodes, wind direction and wind speed, emission intensity and meteorological similarity on the premise of allowing a structural mask, wherein the side weight priors are used for representing the relative influence degree of a next node on another node in a corresponding relation type, and carrying out normalization processing on the side weight priors; the construction and training process of the insulator pollution risk prediction model comprises the following steps: Inputting the multi-dimensional time sequence feature vector to the node feature coding part, mapping the input time sequence feature vector of the heterogeneous node to a hidden space with uniform dimension by the node feature coding part, and outputting node coding representation to the space dependency modeling part; The space dependence modeling part constructs a multi-relation graph structure according to the causal priori information, models the space dependence among nodes through a graph neural network, and obtains updated node space coding representation; Inputting the node space coding representation into the time-dependent modeling part according to a preset time window, and performing time modeling by the time-dependent modeling part to output the comprehensive space-time representation of the node; And inputting the comprehensive space-time representation of the nodes into a fully connected network to obtain the continuous pollution risk index predicted value of the target predicted moment.
2. The method for evaluating and predicting risk of insulator contamination according to claim 1, wherein the multi-dimensional time series feature vector comprises a long-term contamination accumulation feature, a short-term excitation feature and a contamination source related feature; the long-term pollution accumulation characteristics represent the characteristics of long-term pollution accumulation degree and are extracted from the insulator state data; The short-term excitation characteristics represent characteristics of short-term excitation conditions and are extracted from the meteorological environment data; the pollution source characteristic represents the characteristic of the influence of an upstream pollution source and is extracted from the pollution source emission data.
3. The method for evaluating and predicting the risk of contamination of an insulator according to claim 2, wherein the continuous contamination risk index is constructed in the following manner: and carrying out weighted sum and normalized nonlinear transformation on the long-term pollution accumulation characteristic, the short-term excitation characteristic and the pollution source related characteristic to obtain a continuous pollution risk index.
4. The method for evaluating and predicting risk of insulator pollution according to any one of claims 1 to 3, wherein the spatial dependence modeling part uses a graph neural network with a attention mechanism to perform spatial encoding, establishes propagation connection only between node pairs allowed by a structure mask, performs weighted aggregation on information propagation between nodes according to a side weight prior, and obtains a node spatial encoding representation of fused spatial dependence information through a plurality of layers of graph neural networks.
5. The method for evaluating and predicting risk of insulator contamination according to any one of claims 1 to 3, wherein the constraint term is constructed by calculating a deviation between a attention weight and a side weight priori information between nodes learned by a model as a part of a total loss function so as to guide the strength of side influence learned by the model to be consistent with the causal priori information.
6. An insulator contamination risk assessment and prediction system, comprising: The data acquisition and preprocessing unit is configured to acquire multi-source data related to pollution risks of the outdoor insulators and preprocess the multi-source data to construct time-space aligned multi-dimensional time sequence feature vectors, wherein the multi-source data comprises insulator state data, pollution source emission data and meteorological environment data; The index construction unit is configured to construct a continuous pollution risk index for each target node based on the multi-dimensional time sequence feature vector, and the continuous pollution risk index is used as a pollution risk intensity label of the corresponding target node at each moment; The causal priori construction unit is configured to construct causal priori information for describing the action relationship according to the action relationship among pollution source emission, meteorological conditions, plant layout and insulator positions; The model training and updating unit is configured to train a pre-established space-time prediction model by utilizing the multi-dimensional time sequence feature vector and the corresponding continuous pollution risk index to obtain an insulator pollution risk prediction model, wherein the space-time prediction model comprises a node feature coding part, a space dependence modeling part and a time dependence modeling part, the space dependence modeling part models the space dependence among nodes under the constraint of causal prior information, and the time dependence modeling part models the space feature sequence of a target node in a preset time window in time to output a continuous pollution risk index predicted value at a predicted moment; The prediction unit is configured to acquire new multi-source data in real time by utilizing the data acquisition and preprocessing unit, perform preprocessing, update the multi-dimensional time sequence feature vector, input the updated multi-dimensional time sequence feature vector into the insulator pollution risk prediction model and output an insulator pollution risk prediction result; the construction process of the causal priori information comprises the following steps: The method comprises the steps of constructing node types and node sets, namely abstracting related entities in a scene into nodes of different types, forming the node sets, and associating the node types of each node in the node sets, wherein the node types comprise a transformer station/insulator node type, a pollution source node type and a meteorological environment node type; the relation type is constructed by defining a plurality of relation types among nodes to form a relation set according to a physical mechanism and engineering experience, wherein the relation set comprises a pollutant transportation relation, an environment similarity relation and a factory topology or electrical connection relation; constructing a structure mask matrix, namely constructing the structure mask matrix for each relation type in the relation set, wherein elements in the structure mask matrix are used for representing whether directed edges pointing from one node to another node are allowed to exist under the corresponding relation type, if so, the elements are set to be 1, and if not, the elements are set to be 0; The construction of side weight priori information, namely defining side weight priors according to the distance between nodes, wind direction and wind speed, emission intensity and meteorological similarity on the premise of allowing a structural mask, wherein the side weight priors are used for representing the relative influence degree of a next node on another node in a corresponding relation type, and carrying out normalization processing on the side weight priors; the construction and training process of the insulator pollution risk prediction model comprises the following steps: Inputting the multi-dimensional time sequence feature vector to the node feature coding part, mapping the input time sequence feature vector of the heterogeneous node to a hidden space with uniform dimension by the node feature coding part, and outputting node coding representation to the space dependency modeling part; The space dependence modeling part constructs a multi-relation graph structure according to the causal priori information, models the space dependence among nodes through a graph neural network, and obtains updated node space coding representation; Inputting the node space coding representation into the time-dependent modeling part according to a preset time window, and performing time modeling by the time-dependent modeling part to output the comprehensive space-time representation of the node; And inputting the comprehensive space-time representation of the nodes into a fully connected network to obtain the continuous pollution risk index predicted value of the target predicted moment.
7. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the insulator contamination risk assessment and prediction method of any one of claims 1-5 when executing the computer program.
8. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the insulator contamination risk assessment and prediction method of any one of claims 1 to 5.

Description

Insulator pollution risk assessment and prediction method, system, equipment and medium Technical Field The application belongs to the technical field of power system state evaluation and intelligent operation and maintenance, and particularly relates to an insulator pollution risk evaluation and prediction method, system, equipment and medium. Background The outdoor high-voltage insulator is widely applied to transformer substations and power transmission lines, is exposed to a complex atmospheric environment for a long time, is influenced by factors such as industrial emission, dust emission, sea salt, humidity, precipitation and the like, and is easy to form a pollution layer on the surface. Under proper meteorological conditions, the dirty layer absorbs moisture or dissolves, so that the leakage current on the surface of the insulator is obviously increased, the insulation strength is reduced, pollution flashover faults are caused when serious, and equipment tripping and power failure accidents are caused. Therefore, the pollution risk assessment and prediction of the outdoor insulator are important bases for guaranteeing the safe operation of the power grid and implementing the state maintenance based on risks. In the prior art, one common method is to measure pollution quantity indexes such as equivalent salt density (ESDD) and insoluble deposition density (NSDD) through manual sampling and laboratory tests, and perform off-line evaluation on pollution grade and flashover voltage by combining relevant standards or experience curves. The method plays a certain role in insulation design and long-term trend analysis, but has long sampling period and large workload, is difficult to reflect the rapid change of site pollution and meteorological conditions in time, and cannot meet the requirement of online continuous risk assessment. With the development of online monitoring technology, part of transformer substations are provided with leakage current monitoring devices for insulators at typical positions, and attempt to predict leakage current changes by using moving average, ARIMA, gray prediction models and neural network-based time sequence models (such as BP neural network, LSTM and the like), or combine simple thresholds to realize early warning. Such methods have mainly the following problems: (1) Most methods only consider leakage current as a single time sequence for modeling, and cannot fully utilize multi-source information such as pollution source emission, meteorological conditions, environmental background and the like, so that key driving factors in the pollution forming and evolving processes are not fully considered. (2) The prediction target is usually leakage current value or discrete pollution level, and the lack of continuous pollution risk indexes with clear physical meaning and comparison between different sites and working conditions is unfavorable for developing risk assessment and maintenance decision based on uniform scale. (3) The model parameters and the alarm threshold values are set empirically, are sensitive to working conditions such as seasonal changes, running mode changes and the like, have limited stability and generalization capability, and are difficult to continuously provide reliable risk assessment results in complex running environments. In summary, the existing insulator pollution risk assessment method based on traditional measurement and time sequence prediction still has the defects in the aspects of multi-source information fusion, continuous risk characterization, prediction reliability and the like, and is difficult to continuously provide accurate and stable risk assessment results in a complex operation environment. Disclosure of Invention Aiming at the defects existing in the prior art, the application provides an insulator pollution risk assessment and prediction method, system, equipment and medium, which can more fully utilize multi-source data related to the risk of an outdoor insulator to construct pollution risk indexes with definite physical significance, introduce causal prior constraint driven by a mechanism and improve the precision and reliability of insulator pollution risk prediction. The application is realized by the following technical scheme: an insulator contamination risk assessment and prediction method comprises the following steps: Acquiring multi-source data related to pollution risk of an outdoor insulator, preprocessing the multi-source data, and constructing a time-space aligned multi-dimensional time sequence feature vector, wherein the multi-source data comprises insulator state data, pollution source emission data and meteorological environment data; Constructing a continuous pollution risk index for each target node based on the multi-dimensional time sequence feature vector, and taking the continuous pollution risk index as a pollution risk intensity label of the corresponding target node at each moment; constructing causal p