CN-121984236-A - State sensing and fault prediction system and method for low-voltage switch cabinet

CN121984236ACN 121984236 ACN121984236 ACN 121984236ACN-121984236-A

Abstract

The invention discloses a state sensing and fault predicting system and method for a low-voltage switch cabinet, which relate to the technical field of power equipment monitoring and comprise a multisource feature fusion sensing module, a multi-source feature fusion sensing module and a multi-source feature prediction module, wherein the multisource feature fusion sensing module is used for acquiring multisource data of the low-voltage switch cabinet to construct an original state data set, fusing credibility weighting and dynamic and static features and constructing an enhanced state feature vector; the system comprises a health evaluation and fault prediction module, an adaptive control optimization module and a differential control instruction, wherein the health evaluation and fault prediction module is used for generating a health index and a state label through logistic regression mapping based on an enhanced state feature vector, predicting the health index by adopting an attention LSTM, calculating the residual available life, and estimating the fault probability through structural perception k-NN, and the adaptive control optimization module is used for modulating the control advance based on the fault probability to generate the differential control instruction. The invention obviously improves the accuracy of state sensing and the reliability of fault prediction of the low-voltage switch cabinet, and improves the early warning capability and the reliability of operation and maintenance decision.

Inventors

PANG CHANGZHI
LI GUODONG
ZHU SHENGJUN
WANG WENBO

Assignees

杭州华创高科有限公司

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. A state sensing and fault predicting system for a low-voltage switch cabinet is characterized by comprising, The multisource feature fusion perception module is used for acquiring multisource data of the low-voltage switch cabinet to construct an original state data set, fusing reliability weighting and dynamic and static features and constructing an enhanced state feature vector; the health evaluation and fault prediction module is used for generating a health index and a state label through logistic regression mapping based on the enhanced state feature vector, performing health index prediction by adopting attention LSTM, calculating the residual available life, and estimating the fault probability through structural perception k-NN; the self-adaptive control optimization module is used for modulating the control advance based on the fault probability, generating a differentiated control instruction, and acquiring and executing feedback evaluation control efficiency to update and optimize.
2. The system for sensing and predicting the state of the low-voltage switch cabinet is characterized in that the step of collecting multi-source data of the low-voltage switch cabinet to construct an original state data set is characterized in that integrated sensing nodes are deployed at specific physical positions inside the target low-voltage switch cabinet, all the sensing nodes are connected to a local edge computing gateway through an RS485 bus, the edge computing gateway serves as a front-end collecting unit of the controller and is used for collecting data with a fixed period when the power distribution room master clock source is paired, uniform time stamps are given, the multi-source data are aligned to the same moment to form equivalent synchronous data, and the data in the same time window form the original state data set, wherein the original state data set comprises three-phase current effective values, busbar/overlap surface highest temperature, transient high-frequency current waveforms, environment temperature in the cabinet and relative humidity.
3. The system for sensing and predicting the state of the low-voltage switch cabinet is characterized in that the system is used for fusing reliability weighting and dynamic and static characteristics, constructing an enhanced state characteristic vector, namely, based on an original state data set, performing pulse detection processing on a transient high-frequency current waveform to identify partial discharge pulses, calculating apparent discharge capacity in a current characteristic period based on the identified partial discharge pulses, performing differentiation and envelope energy analysis on the transient high-frequency current waveform, identifying effective impact peaks and calculating impact energy concentration, constructing a transient event point set based on the effective impact peaks and performing cluster analysis to obtain a structuring proportion, and constructing a discharge reliability coefficient based on the impact energy concentration and the structuring proportion to perform weighting correction on the apparent discharge capacity to obtain the effective apparent discharge capacity; Based on the effective apparent discharge quantity, the highest temperature of the busbar/overlap surface and the internal environment temperature of the cabinet, respectively obtaining a discharge normalization item and a thermal stress normalization item, respectively fitting to obtain a partial discharge growth slope and a temperature growth slope based on the history data of the effective apparent discharge quantity and the highest temperature of the busbar/overlap surface, further converting the partial discharge growth slope and the temperature growth slope into a partial discharge normalization growth rate and a temperature rise normalization growth rate, fusing the discharge normalization item, the thermal stress normalization item, the partial discharge normalization growth rate and the temperature rise normalization growth rate, constructing a core health feature, namely a coupling degradation factor, taking the imbalance degree of three-phase current and the change rate of humidity as auxiliary features, and finally combining the three-phase current imbalance degree and the change rate of humidity into an enhanced state feature vector, wherein the static feature comprises the discharge normalization item and the thermal stress normalization item, and the dynamic feature comprises the partial discharge normalization growth rate and the temperature rise normalization growth rate.
4. A system for sensing and predicting state of low-voltage switchgear according to claim 3, wherein generating health index and state label by logistic regression mapping based on the enhanced state feature vector calculates linear health score based on the enhanced state feature vector, maps the linear health score to health index, and sets a health lower threshold And an abnormal lower threshold And (2) and When the health index is greater than or equal to the health lower threshold If the health index is equal to or greater than the abnormal lower threshold, the running state is judged to be normal And is less than the lower threshold for health When the health index is smaller than the abnormal lower threshold, the running state is judged to be slightly abnormal And when the running state is judged to be seriously abnormal, finally obtaining the state label.
5. The system for state sensing and fault prediction of a low-voltage switch cabinet according to claim 4, wherein the method is characterized in that attention LSTM is adopted for carrying out health index prediction to calculate remaining usable life, structural sensing k-NN is used for estimating fault probability to store health indexes in a fixed period to form a health index HI time sequence, least square linear regression is carried out on the health index HI time sequence to obtain an average change rate, a global health consistency factor is built as an auxiliary check quantity, a prediction module is triggered based on the average change rate, the global health consistency factor and a state label as judging conditions, the residual usable life estimation is started, an enhanced state feature vector and the health index are built into a current feature vector, fault type label judgment is carried out according to a preset rule base, and fault occurrence probability is calculated.
6. The system for sensing and predicting the state of a low-voltage switch cabinet according to claim 5, wherein the generating differentiated control instructions is based on modulating the control advance based on the fault probability, wherein the generating differentiated control instructions is based on mapping the remaining available life as the basic control advance, and the confidence level modulation is performed on the basic control advance by using the fault probability to obtain the final control advance, and the control instruction set is generated according to the fault type label only when the state label is severely abnormal based on the final control advance.
7. The system for predicting state of low-voltage switch cabinet according to claim 6, wherein said system for optimizing update of control efficiency of acquisition execution feedback evaluation is issued according to control command type, and waits for execution of state feedback, and after receiving valid execution confirmation, the system re-acquires original state data set in fixed period, recalculates to obtain health index after control, and defines control efficiency index, and when the control efficiency index is smaller than control efficiency index threshold, the parameter self-tuning mechanism is triggered to apply updated key parameters to state sensing and fault prediction in next period.
8. A state sensing and fault predicting method for a low-voltage switch cabinet is based on the state sensing and fault predicting system for the low-voltage switch cabinet of any one of claims 1-7, and is characterized by comprising the following steps of, Collecting multi-source data of a low-voltage switch cabinet to construct an original state data set, and fusing reliability weighting and dynamic and static characteristics to construct an enhanced state characteristic vector; based on the enhanced state feature vector, generating a health index and a state label through logistic regression mapping, predicting the health index by adopting an attention LSTM (least squares) to calculate the residual service life, and estimating the fault probability through structural perception k-NN; based on the fault probability modulation control advance, a differential control instruction is generated, the feedback evaluation control efficiency is acquired and executed, and key self-adaptive parameters of state sensing and prediction are updated.
9. A computer device comprising a memory and a processor, said memory storing a computer program, characterized in that said processor, when executing said computer program, implements the steps of a low voltage switchgear state sensing and fault prediction method as claimed in any one of claim 8.
10. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor performs the steps of a method for state sensing and fault prediction of a low voltage switchgear as claimed in any one of the claims 8.

Description

State sensing and fault prediction system and method for low-voltage switch cabinet Technical Field The invention relates to the technical field of power equipment monitoring, in particular to a system and a method for state sensing and fault prediction of a low-voltage switch cabinet. Background In recent years, state sensing and fault prediction technologies of low-voltage switch cabinets gradually evolve from traditional threshold alarming to intelligent and data driving directions. The current mainstream scheme generally adopts a multi-sensor fusion architecture, acquires signals such as partial discharge, temperature, current imbalance and the like, and quantitatively evaluates the state of equipment by constructing a Health Index (HI). Part of advanced systems further introduce machine learning models (such as support vector machines and neural networks) to realize Residual Useful Life (RUL) prediction, and combine a rule engine to perform fault type preliminary discrimination. Meanwhile, the popularization of the edge computing platform enables localization processing of high-frequency transient signals (such as nanosecond discharge pulses) to be possible, and provides a new means for early insulation degradation monitoring. However, despite the increasing complexity of the technical framework, it still faces reliability and practicality bottlenecks in practical engineering applications. Firstly, in a state sensing layer, local discharge signals are extremely easy to be subjected to non-degradation transient interference such as load switching and switching operation, and most systems are only identified by amplitude values or simple statistical characteristics, a multi-dimensional verification mechanism for the physical authenticity of a discharge event is lacked, so that the discharge capacity is high in deficiency and false alarm frequently occurs, secondly, health characteristic construction is limited to current state quantities (such as temperature and discharge capacity), static level and dynamic growth trend cannot be effectively fused, particularly, coupling effect of discharge acceleration and temperature rise deterioration is ignored, early acceleration degradation symptom is difficult to capture, in addition, in a prediction decision layer, residual life prediction often lacks reliable triggering conditions, is usually blindly activated when equipment normally fluctuates, resource waste is caused, a main stream deep learning model output result cannot be verified, interpretability and robustness verification are lacked, whether prediction is reliable cannot be judged, and operation and maintenance personnel cannot make intervention decisions according to the reliability. Therefore, the prior art has the problems that discharge identification is easy to be interfered, degradation dynamic characterization is insufficient, prediction triggering is lack of basis, and result reliability is difficult to guarantee. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a system and a method for state sensing and fault prediction of a low-voltage switch cabinet, which solve the problems that discharge identification is easy to be interfered, degradation dynamic characterization is insufficient, prediction triggering is lack of basis, and result credibility is difficult to guarantee. In order to solve the technical problems, the invention provides the following technical scheme: In a first aspect, the present invention provides a low voltage switchgear state awareness and fault prediction system, comprising, The multisource feature fusion perception module is used for acquiring multisource data of the low-voltage switch cabinet to construct an original state data set, fusing reliability weighting and dynamic and static features and constructing an enhanced state feature vector; the health evaluation and fault prediction module is used for generating a health index and a state label through logistic regression mapping based on the enhanced state feature vector, performing health index prediction by adopting attention LSTM, calculating the residual available life, and estimating the fault probability through structural perception k-NN; the self-adaptive control optimization module is used for modulating the control advance based on the fault probability, generating a differentiated control instruction, and acquiring and executing feedback evaluation control efficiency to update and optimize. The invention relates to a low-voltage switch cabinet state sensing and fault prediction system, which is a preferable scheme, wherein the step of collecting multi-source data of a low-voltage switch cabinet to construct an original state data set is to arrange integrated sensing nodes at specific physical positions inside a target low-voltage switch cabinet, all the sensing nodes are connected to a local e