Search

CN-121997157-A - Intelligent ring main unit fault real-time diagnosis method and system

CN121997157ACN 121997157 ACN121997157 ACN 121997157ACN-121997157-A

Abstract

The invention relates to the technical field of data processing and state monitoring, and discloses a method and a system for diagnosing faults of an intelligent ring main unit in real time, wherein the method comprises the steps of constructing an overcomplete dictionary of fault characteristics; the method comprises the steps of collecting and preprocessing high-frequency operation data streams of a ring main unit in real time, carrying out online sparse decomposition and manifold mapping based on a dictionary, converting high-dimensional signals into low-dimensional sparse coefficients and manifold characteristics, rapidly reasoning by using a lightweight classification model, and triggering fault early warning and data uploading when abnormality is detected. The system comprises a data acquisition module, a preprocessing unit, an edge calculation engine, an intelligent diagnosis module and a communication gateway. The invention realizes the deep dimension reduction and the characteristic enhancement by the manifold sparse coding technology, reduces the bandwidth pressure and the calculation cost on the premise of ensuring the integrity of fault characteristics, and improves the real-time performance and the accuracy of diagnosis.

Inventors

  • ZHOU SHIHAI
  • CHEN LIJING
  • ZHOU MIN

Assignees

  • 湖南旭日电气设备有限公司

Dates

Publication Date
20260508
Application Date
20260409

Claims (10)

  1. 1. The intelligent ring main unit fault real-time diagnosis method is characterized by comprising the following steps of: s1, constructing a fault feature overcomplete dictionary; s2, collecting high-frequency operation data streams of the ring main unit in real time, and preprocessing and sectionally quantizing original signals; S3, performing on-line sparse decomposition and manifold mapping on the real-time high-frequency data stream based on the fault feature overcomplete dictionary, namely circularly executing projection search and signal residual error updating by calculating the inner product similarity of the data frame vector to be processed and each atomic vector in the fault feature overcomplete dictionary until the total energy proportion of residual error energy relative to an original signal is reduced below a preset convergence threshold value, generating a sparse coding vector formed by a non-zero coefficient and a position index of the non-zero coefficient in the dictionary, and then mapping the sparse coding vector to a low-dimensional manifold space which keeps a local topological structure by utilizing manifold transformation to extract the non-intrinsic linear feature reflecting the fault evolution trend; And S4, rapidly reasoning the sparse coefficient by using a lightweight classification model, triggering fault early warning and data uploading when abnormal distribution is detected, namely inputting the extracted low-dimensional manifold features into a preset lightweight classification model for position judgment, monitoring the displacement track of the feature points on the low-dimensional manifold surface and the deviation gradient relative to the normal running centroid in real time, and executing multistage linkage data uploading comprising compression feature coefficients or reconstructed original waveform fragments according to the fault severity level when the judgment result meets the preset risk region condition.
  2. 2. The method for diagnosing faults of the intelligent ring main unit in real time according to claim 1, wherein the construction of the fault feature overcomplete dictionary comprises the steps of obtaining multisource fault waveform samples of the intelligent ring main unit in a historical operation period and performing physical range standardization processing, performing direction iterative adjustment on randomly generated base vectors by utilizing a residual minimization principle, extracting atomic features representing essential physical attributes of different fault types from the base vectors, and orderly arranging the extracted atomic features according to feature types and physical directivities to form the fault feature overcomplete dictionary with the capability of sparse representation of typical fault features.
  3. 3. The method for diagnosing faults of the intelligent ring main unit according to claim 2, wherein the step S2 comprises the steps of obtaining multi-source analog signals through sensors arranged on key nodes of the ring main unit, performing anti-aliasing filtering processing to filter high-frequency clutter interference, performing high-frequency discretization conversion, dividing a generated sampling sequence into continuous data frames according to a fixed data scale, and simultaneously performing precision adjustment and segmentation encapsulation on the data frames according to real-time available capacity of an edge side memory buffer area.
  4. 4. The method for diagnosing faults of the intelligent ring main unit in real time according to claim 3, wherein the step S1 specifically comprises the steps of obtaining arc fault waveforms, partial discharge characteristic fingerprints and mechanical vibration characteristic sequences of the intelligent ring main unit in a historical operation period; Performing physical range normalization processing on various fault waveform samples, converting analog quantities acquired by a sensor into floating point number sequences in a fixed numerical value interval, and performing phase alignment processing on the similar fault samples through positioning signal initial trigger points to construct an original training sample set; And carrying out iterative optimization on the original training sample set through an offline learning algorithm, searching a parameter set which represents an original waveform by a minimum number of basis vector combinations in each iteration, extracting atomic characteristics which represent arc transient spike, partial discharge pulse oscillation and mechanical action impact energy distribution, and solidifying and storing the atomic characteristics in a storage unit of edge side equipment.
  5. 5. The method for diagnosing faults of the intelligent ring main unit according to claim 4, wherein the step S2 specifically comprises the steps of performing anti-aliasing filtering on the analog electric signals collected by the sensor by utilizing a low-pass Butterworth filter to filter out high-frequency components higher than half of the sampling frequency; Analog-to-digital conversion is carried out on the analog signal according to a preset sampling rate, so that microsecond-level partial discharge pulse envelope is ensured to be captured; cutting and packaging the continuously generated sampling point sequence according to a preset length, dynamically adjusting the quantization bit depth according to the real-time load of the processor after the data frame enters the memory buffer area, and reserving high-precision storage for the signal segment with severe waveform change.
  6. 6. The real-time fault diagnosis method of intelligent ring main unit according to claim 5, wherein the specific logic of the online sparse decomposition in step S3 is that a matching pursuit algorithm is adopted to calculate the inner product of the signal vector of the current data frame and each atomic vector in the overcomplete dictionary of the fault characteristics, and the atom with the largest inner product value is searched out as the current main component; subtracting the projection component of the main component in the atomic direction from the current signal to obtain a residual signal, and continuously searching atoms with the strongest correlation with the residual signal in a residual atom library; And repeatedly performing the processes of projection, subtraction and re-optimizing until the residual energy ratio is lower than a preset proportion or the number of atoms selected reaches a preset upper limit, and converting the high-dimensional data frame into sparse features containing a small number of non-zero values.
  7. 7. The real-time diagnosis method of intelligent ring main unit fault according to claim 6, wherein the specific logic of manifold mapping in step S3 is that the sparse coding vector generated by local linear embedding or equidistant mapping mechanism is utilized to construct a neighbor graph reflecting global eigenvector by calculating local euclidean distance between high-dimensional feature points; Searching for a coordinate embedded representation for maintaining the local topological relation of the neighbor graph in a low-dimensional Euclidean space, and expanding fault characteristics of nonlinear bending distribution in the low-dimensional space; And utilizing the topological consistency of manifold space to gather and strengthen the fault characteristics of the same type and execute nonlinear decoupling on fault modes of different types.
  8. 8. The method for diagnosing faults of the intelligent ring main unit according to claim 7, wherein the specific logic for triggering fault early warning and data uploading in the step S4 is that a fault early warning signal is triggered when the lightweight classification model judges that characteristic points fall on the periphery of a normal operation area and the deviation speed gradient exceeds a preset threshold value; when the judging result indicates a suspected fault state, the communication gateway preferentially builds a data packet containing the current fault index, the sparse coefficient value and the fault moment and sends the data packet to the management platform; and when the judging result indicates a serious fault or insulation breakdown state, the original sampling data frame in the memory is called, waveform reconstruction is carried out by using the sparse coding vector, and the reconstructed high-fidelity fault waveform segment is uploaded preferentially.
  9. 9. The intelligent ring main unit fault real-time diagnosis method according to claim 8, further comprising a dictionary self-adaptive updating step, wherein an edge calculation engine monitors residual energy distribution in a sparse decomposition process in real time and calculates a data duty ratio of residual energy which is continuously higher than the preset proportion of original signal energy in a preset monitoring period; When the data duty ratio exceeds a preset frequency threshold, automatically starting a background updating task by the system, marking a data segment with abnormal residual energy as an unknown characteristic sample, and uploading the unknown characteristic sample to the cloud; and the cloud management platform performs cluster analysis on the unknown characteristic samples by using an offline learning algorithm, generates patch atoms and transmits the patch atoms to the edge computing engine through the communication gateway.
  10. 10. The intelligent ring main unit fault real-time diagnosis system for realizing the method of any one of claims 1 to 9 is characterized by comprising a data acquisition module, a control module and a control module, wherein the data acquisition module is configured to acquire electric signals, vibration signals and environmental parameters of key nodes in the ring main unit; the preprocessing unit is connected with the data acquisition module and is configured to perform anti-aliasing filtering, high-frequency sampling and segmented packaging on the original signal, and dynamically adjust the quantization precision according to the edge side resource state; The edge computing engine is internally preset with a fault characteristic overcomplete dictionary and is configured to receive the preprocessed data frame, perform on-line sparse decomposition and manifold characteristic mapping based on a matching pursuit algorithm and output sparse coding vectors and low-dimensional manifold intrinsic characteristics which are formed by non-zero coefficients; The intelligent diagnosis module is connected with the edge calculation engine and is configured to classify and judge the low-dimensional manifold features based on the lightweight classification model, output fault state labels through the monitoring feature displacement track, and the communication gateway is configured to execute bandwidth self-adaptive adjustment and selectively send compressed feature data or reconstructed high-fidelity waveform information to the remote monitoring center according to the fault state labels output by the intelligent diagnosis module.

Description

Intelligent ring main unit fault real-time diagnosis method and system Technical Field The invention belongs to the field of data processing and state monitoring, and particularly relates to a method and a system for diagnosing faults of an intelligent ring main unit in real time. Background Along with the acceleration of intelligent power distribution network construction, intelligent looped netowrk cabinet is as core node and key component part of distribution system, and its running state's real-time supervision and intelligent fortune dimension level directly relate to holistic security and the reliability of electric wire netting, and circuit breaker, load switch, mutual-inductor and all kinds of environmental monitoring sensors have been integrated to looped netowrk cabinet inside, and the steady operation of these subassemblies is the important basis of guaranteeing distribution network and continuously supplying power and high-efficient dispatch. The fault diagnosis technology is a core means for improving the operation and maintenance intelligent degree of the ring main unit, and aims to realize early discovery and accurate positioning of potential defects of equipment, effectively reduce the risk of unplanned power failure and prolong the service life of power distribution equipment by collecting and logically analyzing electric parameters, environmental parameters and high-frequency characteristic signals generated in the operation process of the ring main unit in real time. The traditional ring main unit fault diagnosis technology mainly has the following defects that firstly, along with continuous improvement of sampling frequency, the data scale generated by high-frequency partial discharge signals and vibration waveforms is exponentially increased, so that the pressure of data acquisition and storage is huge, secondly, the real-time performance and transmission bandwidth of data processing are contradictory, if the full quantity of original data are uploaded to a cloud end, serious communication delay and bandwidth consumption are generated, if simple frequency reduction or mean value compression processing is carried out on the edge side, key waveform details reflecting fault essence are lost, so that the diagnosis accuracy is greatly reduced, and further, the edge side computing resources are relatively limited, a high-efficiency algorithm capable of simultaneously realizing mass data dimension reduction and complex feature enhancement is not easy to complete under the limited calculation force condition, finally, the characterization capability of a system on nonlinear and non-stable fault signals is not enough, false alarm or missing alarm is extremely easy to generate under the complex operation environment, the requirements of an intelligent power grid on fault millisecond response and accurate judgment are difficult to meet, and the problems limit the application effect of the intelligent ring main unit fault diagnosis system in the actual power production environment. Disclosure of Invention The invention aims to provide a method and a system for diagnosing faults of an intelligent ring main unit in real time, which can effectively solve the problems in the background technology. Aiming at the core contradiction between the explosion of the data volume generated by the intelligent ring main unit under the high-frequency sampling environment and the limited transmission bandwidth and the diagnosis difficulty that the real-time performance and the detail characteristic reservation can not be achieved due to the limitation of the computing resources on the edge side of the original data, the invention realizes the deep dimension reduction and the light weight real-time diagnosis of the mass high-frequency data on the premise of ensuring the integrity of the fault characteristics by introducing the manifold sparse coding technology. The invention provides a real-time fault diagnosis method for an intelligent ring main unit, which comprises the following steps of S1, constructing an overcomplete dictionary of fault characteristics, S2, collecting high-frequency operation data streams of the ring main unit in real time, preprocessing and sectionally quantizing original signals, S3, carrying out online sparse decomposition and manifold mapping on the real-time high-frequency data streams based on the overcomplete dictionary of the fault characteristics, converting the high-dimensional original signals into sparse coefficients and manifold characteristics in a low-dimensional space, S4, carrying out quick reasoning on the sparse coefficients by utilizing a lightweight classification model, and triggering fault early warning and data uploading when abnormal distribution is detected. The method comprises the steps of S11, obtaining multi-source fault waveform samples of an intelligent ring main unit in a historical operation period, wherein the multi-source fault waveform samples comprise arc f