CN-121997097-A - Insulator surface pollution prediction method based on meteorological environment factors

CN121997097ACN 121997097 ACN121997097 ACN 121997097ACN-121997097-A

Abstract

The application relates to the technical field of overhead transmission lines and discloses a method for predicting surface pollution of an insulator based on meteorological environment factors, which comprises the steps of firstly obtaining multi-source heterogeneous data of insulator body attributes, microclimate environment and atmospheric chemical components and constructing a historical sample database; the method comprises the steps of obtaining a normalized time sequence matrix through data cleaning and feature screening, constructing a mixed neural network model, extracting local features through a convolutional neural network, generating an attention mask based on relative humidity by utilizing a dynamic weighting chemical component feature fusion network, dynamically weighting feature vectors, inputting the feature vectors into a long-period and short-period memory network for time sequence evolution prediction, and finally outputting predicted values of equivalent salt density and insoluble deposition density on the surface of an insulator by utilizing a training convergence model. According to the application, by introducing a humidity attention mechanism, the deep fusion of the meteorological environment and the chemical component characteristics is realized, and the prediction precision of the insulator dirt accumulation in the complex environment is remarkably improved.

Inventors

WANG XIAOQING
HUANG JUNLAN
QIN HAODONG
ZHOU JIAYI
LI WENRONG
LIAO YUQIN
LIU XIAOBING
CHANG AN
WEI XIAOXING

Assignees

中国南方电网有限责任公司超高压输电公司电力科研院

Dates

Publication Date: 20260508
Application Date: 20251229

Claims (10)

1. The insulator surface pollution prediction method based on meteorological environment factors is characterized by comprising the following steps of: Step one, multi-source heterogeneous data related to accumulation of the surface pollution of the insulator are obtained, wherein the multi-source heterogeneous data comprise insulator body attribute data, microclimate environment data and atmospheric chemical composition data, and corresponding actual measurement values of the surface pollution degree of the insulator are obtained; Step two, carrying out data cleaning, normalization processing and feature screening on the multi-source heterogeneous data in the historical sample database to obtain a normalized multi-variable time sequence matrix; Step three, constructing a mixed neural network model based on the environmental weather mutual attention and a long-short-term memory network of the convolutional neural network, wherein the mixed neural network model comprises a convolutional neural network local feature extraction network, a dynamic weighting chemical component feature fusion network and a long-short-term memory network time sequence evolution prediction network which are connected in sequence; Training the hybrid neural network model by using a back propagation algorithm based on data in a historical sample database until the model converges; Fifthly, predicting the surface pollution degree of the insulator in the period to be predicted by using the trained hybrid neural network model, and outputting a prediction result.
2. The method for predicting surface pollution of an insulator based on meteorological environment factors according to claim 1, wherein in the first step, the insulator body attribute data includes insulator structure height, nominal diameter, umbrella protrusion, umbrella spacing, and insulator material type; The microclimate environment data comprise environment temperature, relative humidity, wind speed, wind direction and rainfall; the atmospheric chemical composition data comprises a particulate matter concentration with an aerodynamic diameter of less than or equal to 2.5 microns, a particulate matter concentration with an aerodynamic diameter of less than or equal to 10 microns, a sulfur dioxide concentration, a nitrogen dioxide concentration, a carbon monoxide concentration and an ozone concentration; The measured values of the surface pollution degree of the insulator comprise equivalent salt density and insoluble deposition density.
3. The method for predicting surface contamination of an insulator based on meteorological environmental factors of claim 1, wherein in the step two, the step of cleaning the data comprises: taking the actual measurement value of the pollution degree on the surface of the insulator as a target variable, and taking microclimate environment data and atmospheric chemical component data as characteristic variables to construct a decision tree model; calculating the depth of each leaf node and the proportion of samples contained in the node by traversing all leaf nodes of the decision tree model; Marking leaf nodes with depth smaller than a preset depth threshold or sample proportion smaller than a preset proportion threshold as abnormal nodes; And eliminating all sample data falling into the abnormal node from a historical sample database.
4. The method for predicting surface pollution of an insulator based on meteorological environment factors according to claim 1, wherein in the step two, the step of feature screening includes: Adopting a gray correlation analysis method, defining a sequence of actual measurement values of the surface pollution degree of the insulator as a reference sequence, and defining a sequence of normalized insulator body attribute data, microclimate environment data and atmospheric chemical component data as a comparison sequence; calculating gray correlation degree between each comparison sequence and the reference sequence through a gray correlation analysis algorithm; And reserving characteristic indexes with gray correlation degree larger than a preset correlation degree threshold, and eliminating characteristic indexes with gray correlation degree smaller than or equal to the preset correlation degree threshold.
5. The method for predicting the surface pollution of the insulator based on the meteorological environment factors according to claim 1, wherein in the third step, the local feature extraction network of the convolutional neural network comprises a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and a flattening layer which are sequentially connected; Inputting the normalized multivariable time sequence matrix into the first convolution layer, performing two-layer convolution operation and maximum pooling operation, and performing operation of unfolding by the flattening layer to obtain a one-dimensional high-dimensional feature vector by calculation; wherein the dynamically weighted chemical component feature fusion network performs the following operations: And (3) receiving the high-dimensional feature vector output by the local feature extraction network of the convolutional neural network and the relative humidity data processed in the second step, generating an attention mask by using the relative humidity data, weighting the high-dimensional feature vector by using the attention mask to obtain a fusion feature vector, and inputting the fusion feature vector into the long-term and short-term memory network time sequence evolution prediction network.
6. The method for predicting surface pollution of an insulator based on meteorological environment factors of claim 5, wherein in the third step, the specific operation process of the dynamic weighted chemical component feature fusion network comprises: acquiring a relative humidity normalization value of the current time step; Inputting the relative humidity normalized value to a full-connection layer, and mapping scalar-form humidity data into a humidity characteristic vector through matrix multiplication calculation, wherein the dimension of the humidity characteristic vector is the same as that of the high-dimensional characteristic vector; processing the humidity characteristic vector through a Sigmoid activation function to generate a humidity attention mask vector with a value between 0 and 1; the fusion feature vector is obtained by performing element-by-element multiplication calculation on the high-dimensional feature vector and the humidity attention mask vector.
7. The method for predicting the surface pollution of the insulator based on the meteorological environment factors according to claim 6, wherein in the third step, the long-period memory network time sequence evolution prediction network is composed of a plurality of long-period memory network storage units which are unfolded in time sequence; Each long-short-term memory network storage unit receives the fusion feature vector at the current moment and the hidden layer state at the last moment as inputs; The memory unit calculates forgetting information through a forgetting gate, calculates cell state update information through an input gate and calculates hidden layer output at the current moment through an output gate; And the hidden layer output of the last time step is connected to a full-connection regression layer, and the full-connection regression layer outputs a normalized predicted value of the surface pollution degree of the insulator.
8. The method for predicting surface contamination of an insulator based on meteorological environmental factors of claim 1, wherein in step four, the training strategy comprises: Taking the mean square error as a loss function, and calculating the error between the normalized predicted value output by the model and the normalized value of the actual measurement value of the surface pollution degree of the insulator through the loss function; adopting an adaptive moment estimation optimization algorithm, calculating first moment estimation and second moment estimation of the gradient through an adaptive moment estimation formula based on the gradient of the loss function relative to network parameters, and adaptively updating a weight matrix and a bias vector in a network; and introducing a random inactivation regularization strategy between the long-period memory network time sequence evolution prediction network and the full-connection regression layer, and randomly setting the activation value of part of neurons to be zero in training iteration.
9. The method for predicting the surface pollution of the insulator based on the meteorological environment factors according to claim 8, wherein the loss function, namely, the mean square error, is calculated in the following manner: obtaining a model predicted value and a corresponding actually measured tag value of each sample in a training batch; Calculating the square of the Euclidean distance between the model predicted value and the actually measured tag value of each sample through the Euclidean distance formula; And calculating the loss value of the current batch through the operation of summing the square errors of all samples in the training batch and taking an average value.
10. The method for predicting surface pollution of an insulator based on meteorological environment factors as set forth in claim 1, wherein in the fifth step, the step of outputting the prediction result includes: Obtaining the maximum value and the minimum value of the actual measurement value of the pollution degree on the surface of the insulator in the training set data; performing inverse normalization linear inverse transformation calculation on the normalized predicted value output by the hybrid neural network model by utilizing the maximum value and the minimum value to obtain a real predicted value with physical dimension; the true predictions include milligrams per square centimeter equivalent salt density predictions and insoluble deposition density predictions.

Description

Insulator surface pollution prediction method based on meteorological environment factors Technical Field The invention relates to the technical field of overhead transmission lines, in particular to a method for predicting pollution on the surface of an insulator based on meteorological environment factors. Background The insulator is used as a key external insulation component in power transmission and transformation lines and distribution systems, and the running state of the insulator is directly related to the safety stability of the power system. During long-term operation, a pollution layer consisting of industrial dust, salt mist and the like is gradually accumulated on the surface of the insulator. When the wet weather is met, the pollution layers can be dissolved to form a conductive water film, so that the surface leakage current of the insulator is increased sharply, the dry flash voltage is reduced, and serious power accidents such as pollution flashover, tripping and the like are caused. With the continuous expansion of the power grid scale and the extension of lines to coastal, heavy industrial areas, wind and sand areas and other complex environments, the problem of pollution on the surface of the insulator is more and more remarkable, and the method provides a serious challenge for the operation and maintenance of the power equipment. The accumulation of insulator pollution is a complex dynamic process modulated by multidimensional factors, which is not only dependent on the concentration of suspended particles and chemical gases in the atmosphere, but also is closely related to the microclimate conditions such as ambient temperature, relative humidity, rainfall, wind speed and direction. There are complex nonlinear coupling effects between these factors, for example, relative humidity not only affects the wetting degree of the contaminant particles, but also changes the deliquescence properties and adhesion of the chemical components, thereby significantly affecting the deposition rate. However, traditional insulator contamination level assessment relies mainly on manual periodic tower climbing sampling detection or based on simple point monitoring equipment to acquire local data. The method has obvious hysteresis, is difficult to reflect the trend of the pollution dynamic change along with time in real time, has high manual detection cost and long period, and cannot meet the requirement of intelligent operation and maintenance of the power grid. Although the development of internet of things and online monitoring technology in recent years makes it possible to acquire massive weather and environment data, the existing pollution prediction technology still has significant limitations. The traditional mathematical statistical regression or shallow machine learning algorithm is mostly based on the current prediction model, and the method usually regards microclimate parameters and atmospheric chemical components as parallel characteristic input independent of each other when modeling, ignores deep interaction mechanisms between different physical fields, and particularly cannot effectively describe nonlinear modulation effect of relative humidity on deposition efficiency of specific chemical components. In addition, the pollution accumulation process has long-period time sequence dependency, and the conventional model is difficult to simultaneously consider the space feature extraction of multi-source heterogeneous data and the evolution law capture of long time span, so that the prediction precision of equivalent salt density (ESDD) and insoluble deposition density (NSDD) on the surface of an insulator is difficult to meet the actual requirements of refined differential operation and maintenance under complex and changeable meteorological conditions. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a method for predicting the surface pollution of an insulator based on meteorological environment factors, which solves the problem that the prediction precision is insufficient under complex meteorological conditions due to the fact that nonlinear strong coupling relation between large-atmosphere chemical components and micro-meteorological environments cannot be fully considered in the conventional transmission line pollution degree prediction technology. The method of the embodiment of the invention firstly constructs a multi-source heterogeneous database containing insulator body attributes, microclimate environment and atmospheric chemical components, and aligns all data sources in time dimension to form a historical sample set for supervised learning. Aiming at noise interference and redundant characteristics existing in original monitoring data, the method generates a high-quality normalized multi-variable time sequence matrix through a data cleaning and characteristic screening strategy, and provides a standardized data base for model input. On the co