CN-122026610-A - Electrical state intelligent monitoring method for distribution cable branch box

CN122026610ACN 122026610 ACN122026610 ACN 122026610ACN-122026610-A

Abstract

The invention belongs to the field of power equipment monitoring, and particularly relates to an intelligent electrical state monitoring method for a distribution cable branch box. The method comprises the steps of obtaining state monitoring signals and in-box environment signals of all electrical connection points in a branch box to be detected, preprocessing the state monitoring signals and the in-box environment signals to obtain standardized state parameters, fusing the state parameters and preset reference data in a cross-time scale mode to construct a multi-dimensional feature vector, inputting the multi-dimensional feature vector into a preset depth diagnosis network to obtain diagnosis results and confidence degrees of potential fault types, and determining fault monitoring results of the branch box to be detected according to the diagnosis results and the confidence degrees. The invention obviously improves the accuracy and the signal-to-noise ratio of the monitoring data, thereby reducing the risks of false alarm and missing alarm of faults.

Inventors

Xu Qianquan
CHEN SI
Chen Aie
ZHENG YUANFEI

Assignees

浙江飞沃电气有限公司

Dates

Publication Date: 20260512
Application Date: 20260119

Claims (9)

1. An intelligent electrical state monitoring method for a distribution cable branch box is characterized by comprising the following steps: Acquiring state monitoring signals and in-box environment signals of all electrical connection points in a branch box to be tested; Preprocessing the state monitoring signal and the in-box environment signal to obtain standardized state parameters so as to eliminate interference of environment variables on the state monitoring signal; Fusing the state parameters and preset reference data in a cross-time scale manner to construct a multi-dimensional feature vector, wherein the reference data are used for indicating the historical context and the environmental change trend of the operation of the branch box, and the multi-dimensional feature vector is used for representing the health state of each electrical connection point; Inputting the multi-dimensional feature vector into a preset depth diagnosis network to obtain a diagnosis result and a confidence level of potential fault types, wherein the depth diagnosis network integrates an attention mechanism layer and a feature weighting module, the attention mechanism layer is used for evaluating the contribution weight of each basic feature element in the multi-dimensional feature vector, and the feature weighting module is used for selectively enhancing and fusing the corresponding basic feature elements according to the contribution weight so as to generate comprehensive features for diagnosis; and determining a fault monitoring result of the branch box to be tested according to the diagnosis result and the confidence coefficient.
2. The method of claim 1, wherein the condition monitoring signal comprises a temperature signal, a partial discharge ultrasonic signal, and a transient voltage-to-ground signal, the preprocessing comprising a time domain alignment process comprising: taking a power frequency signal of the power grid analyzed from the state monitoring signal as a reference clock; determining a reference time scale sequence according to the period and the phase of the reference clock; And according to the reference time scale sequence, synchronizing the original data sequences of the temperature signals, the partial discharge ultrasonic signals and the transient voltage-to-ground signals of all the electrical connection points, and simultaneously, according to the reference time scale sequence, carrying out interpolation resampling on the original data sequences of the in-box environment signals so as to enable the original data sequences of all the signals to reach time reference alignment.
3. The method of claim 1, wherein the preprocessing further comprises a coupling compensation process, the coupling compensation process comprising: acquiring temperature and humidity data of the branch box to be tested; according to the temperature and humidity data, constructing a dynamic interference model acting on the state monitoring signal; using the dynamic interference model to eliminate the additional temperature rise of the environment for the temperature signal in the state monitoring signal; performing sound velocity correction processing based on the ambient temperature on the propagation delay of the partial discharge ultrasonic signal in the state monitoring signal by using the dynamic interference model; And carrying out self-adaptive adjustment processing on the detection threshold value and the gain of the transient voltage signal to ground in the state monitoring signal by utilizing the dynamic interference model.
4. The method according to claim 1, wherein the construction method of the reference data comprises: Continuously storing historical operation data of the branch box to be tested, forming a historical database containing a load current time sequence curve and event records, and acquiring weather forecast data of a future preset period from an authorized weather service interface according to a preset period; And carrying out structural association and indexing on the historical database and the weather forecast data according to a unified time axis to form a reference data set, and extracting corresponding historical working condition fragments from the reference data set according to the time stamp of the real-time state parameter in the cross-time scale fusion process to carry out fusion calculation on the historical working condition fragments and the future environment prediction sequence.
5. The method according to claim 1, wherein the method for constructing the multi-dimensional feature vector comprises: extracting parallel characteristics of various state parameters, wherein the parallel characteristics extraction comprises the steps of extracting time domain trend and inter-phase difference characteristics from a temperature signal, extracting statistical spectrum, phase distribution and pulse waveform characteristics from a partial discharge signal, and extracting effective values, pulse counts and frequency domain characteristics from a transient voltage signal to ground; Organizing the features obtained by extracting the parallel features according to a preset sliding time window, and determining statistics in the windows and time sequence evolution features among the windows; matching and performing association analysis on the time sequence evolution characteristics, historical load modes of corresponding time periods in the reference data and meteorological conditions, and determining context derivative characteristics; And normalizing and vector splicing the features obtained by extracting the parallel features, the time sequence evolution features and the context derivative features to form the multidimensional feature vector.
6. The method of claim 1, wherein the deep diagnostic network is further integrated with a classification output layer, the diagnostic process of the deep diagnostic network comprising: Through the attention mechanism layer, carrying out dynamic weight evaluation on each basic feature element in the multidimensional feature vector; selectively enhancing and fusing the basic feature elements according to the dynamic weights through the feature weighting module to generate comprehensive features; inputting the comprehensive characteristics to a classification output layer of the depth diagnosis network, and mapping the comprehensive characteristics into probability values corresponding to potential fault categories through the classification output layer; and outputting the fault category with the highest probability value as the diagnosis result, and outputting the highest probability value as the confidence level.
7. The method of claim 1, wherein determining a fault monitoring result for the branch box to be tested according to the diagnosis result and the confidence comprises: Mapping the diagnosis result to a predefined fault type, and simultaneously comparing the confidence coefficient with preset high, medium and low threshold intervals to determine the confidence level; Determining a comprehensive risk level according to the fault type and the confidence level; And outputting the comprehensive risk level as a fault monitoring result of the branch box to be tested.
8. The method of claim 1, wherein obtaining a status monitor signal for each electrical connection point in the branch box to be tested comprises: Collecting surface temperature signals of all the electrical connection points through a contact type or infrared temperature sensor; Collecting acoustic emission signals generated by partial discharge through an ultrasonic sensor array arranged in the box body; Electromagnetic signals related to the insulation state are collected through a high-frequency current transformer or a transient voltage-to-ground sensor coupled to a ground loop.
9. The method of claim 1, wherein obtaining an in-box environmental signal within the branch box to be tested comprises: Acquiring environmental temperature and relative humidity data through a temperature and humidity sensor arranged in the box body; the open/close state signal of the box door is collected by a door magnetic sensor or a micro switch installed at the box door.

Description

Electrical state intelligent monitoring method for distribution cable branch box Technical Field The invention belongs to the technical field of power equipment monitoring, and particularly relates to an intelligent electrical state monitoring method for a distribution cable branch box. Background Distribution cable trunks are critical node equipment in a distribution network system that internally contains a plurality of cable connectors and electrical connection points. In the long-term operation process of the connection points, the problems of poor contact, insulation aging and the like are easy to occur under the combined action of power, heat and mechanical stress, so that overheating, partial discharge and even short-circuit faults are caused, and the reliability and the safety of power supply are seriously threatened. Therefore, the method effectively monitors the running state of the electrical connection point in the branch box, and early warns potential faults in time, thereby having important engineering value. The existing monitoring method is to monitor signals such as temperature, partial discharge ultrasound or transient voltage to ground. However, the methods have the defects that on one hand, the monitoring signals are extremely easy to be interfered by dynamic environment variables such as temperature and humidity of the environment inside and outside the box, load current fluctuation and the like, for example, the environment temperature can be directly superposed on the temperature rise of a connecting point, the humidity can influence the propagation of ultrasonic signals and the discharge threshold value of an insulating surface, the interference factors lead to the distortion of the monitoring data and are easy to generate false alarm or missing alarm, on the other hand, the traditional diagnosis method generally carries out threshold value judgment based on instantaneous data, the comprehensive consideration of the historical context (such as a historical load mode) of equipment and the future environment change trend (such as future weather) is lacked, the evolution rule of faults is difficult to be accurately grasped from a longer time scale, and the predictability and the accuracy of diagnosis are limited. In addition, when multi-source information such as temperature, partial discharge, voltage and the like is processed, the related technologies are often fused by adopting simple weighting or logic judgment, the contribution weights of different features in fault diagnosis cannot be dynamically evaluated according to real-time working conditions, deep symptoms which are strongly related to a specific fault mode in multi-dimensional data are difficult to effectively mine, and the reliability of a diagnosis result is influenced. Disclosure of Invention In order to solve the problems in the prior art, namely the problems that monitoring signals are easy to be interfered by environment, diagnosis lacks time sequence context association and multi-dimensional feature fusion diagnosis precision is low in the prior art, the invention provides an intelligent electrical state monitoring method for a distribution cable branch box, which is characterized by comprising the following steps: Acquiring state monitoring signals and in-box environment signals of all electrical connection points in a branch box to be tested; Preprocessing the state monitoring signal and the in-box environment signal to obtain standardized state parameters so as to eliminate interference of environment variables on the state monitoring signal; Fusing the state parameters and preset reference data in a cross-time scale manner to construct a multi-dimensional feature vector, wherein the reference data are used for indicating the historical context and the environmental change trend of the operation of the branch box, and the multi-dimensional feature vector is used for representing the health state of each electrical connection point; Inputting the multi-dimensional feature vector into a preset depth diagnosis network to obtain a diagnosis result and a confidence level of potential fault types, wherein the depth diagnosis network integrates an attention mechanism layer and a feature weighting module, the attention mechanism layer is used for evaluating the contribution weight of each basic feature element in the multi-dimensional feature vector, and the feature weighting module is used for selectively enhancing and fusing the corresponding basic feature elements according to the contribution weight so as to generate comprehensive features for diagnosis; and determining a fault monitoring result of the branch box to be tested according to the diagnosis result and the confidence coefficient. In some preferred embodiments, the condition monitoring signal comprises a temperature signal, a partial discharge ultrasonic signal, and a transient voltage-to-ground signal, the preprocessing comprises a time domain alignment pro