CN-121997248-A - Time domain alternating current equivalent direct current resistance test and analysis method based on big data

CN121997248ACN 121997248 ACN121997248 ACN 121997248ACN-121997248-A

Abstract

The invention relates to the technical field of power equipment detection and state evaluation, in particular to a time domain alternating current equivalent direct current resistance test and analysis method based on big data; according to the invention, the equivalent direct current resistance is calculated through iterative optimization of a linear prediction model and a time domain signal, the equivalent resistance time sequence characteristic and the working condition related characteristic are fused, a high-dimensional characteristic matrix is constructed, the intrinsic law and the working condition influence mechanism of data are fully excavated, rich characteristic support is provided for the model, the basic characteristic classification is good for the model through a geometric cascade forest model, the health state evaluation and the anomaly detection are considered by a gradient lifting tree and an isolated forest fusion model, the two are complemented to form double research judgment, the comprehensiveness and the accuracy of state identification are improved, the problem of the data or the model is found in time through comparison of a double prediction result and output of an anomaly feedback list, the direction is provided for subsequent optimization, and the stability and the reliability of an analysis result are ensured.

Inventors

ZHANG LEI
Wang Handie
CHEN RONGYU
GAO DAYONG
LIU ZHENGQIAN
HE WEIHANG
JI YANG
ZHENG HAI
SONG JINGRU

Assignees

国网辽宁省电力有限公司鞍山供电公司

Dates

Publication Date: 20260508
Application Date: 20251224

Claims (8)

1. The time domain alternating current equivalent direct current resistance testing and analyzing method based on big data is characterized by comprising the following steps: The method comprises the steps of firstly, calculating an equivalent direct current resistance through a dual path, namely, collecting sample data to construct a prediction model, and calculating in real time to obtain an equivalent direct current resistance ZR; the validity of the resistor is checked, namely, the deviation between ZR and RD is calculated, and an effective signal is generated when the deviation is within a preset range, otherwise, an ineffective signal is generated and early warning is triggered; step three, constructing characteristics and storage data, namely after effective signals are generated, associating storage resistance, time and working condition data, and simultaneously fusing resistance time sequence characteristics and working condition association characteristics to construct a high-dimension characteristic matrix and a basic characteristic vector; Inputting the basic feature vector into a geometric cascading forest model to obtain a first prediction result, inputting the high-dimensional feature matrix into a gradient lifting tree and isolated forest fusion model, and combining with interactive analysis to obtain a second prediction result; And fifthly, verifying and feeding back the results, namely comparing the two prediction results, outputting the results if the two prediction results are consistent, generating a deviation signal if the two prediction results are inconsistent, and constructing an abnormal feedback list.
2. The method for testing and analyzing the time-domain alternating-current equivalent direct-current resistance based on big data according to claim 1, wherein the equivalent direct-current resistance ZR is obtained by the following steps: s1, collecting alternating-current time domain waveform data of n (n > 5) groups of power equipment as sample data, and extracting two core variables (a predicted variable and a reference variable) in the sample data; carrying out standardization processing on the predicted variable, and carrying out iteration processing on the predicted variable and the reference variable which are subjected to standardization processing through a set least square method to obtain a final coefficient vector beta; S2, acquiring alternating current time domain waveform data of m (m > 0) groups of power equipment in real time, and carrying out standardization processing on the alternating current time domain waveform data of each group to obtain a prediction variable Xnew; s3, calculating to obtain equivalent direct current resistance values of each group based on R=prediction variable Xnew×final coefficient vector beta; And S4, carrying out average value calculation on the equivalent direct current resistance value, and taking the average value calculation result of the equivalent direct current resistance value as a final equivalent direct current resistance ZR.
3. The method for testing and analyzing time-domain ac equivalent dc resistance based on big data according to claim 1, wherein the process of obtaining the final equivalent resistance RD is as follows: the method comprises the steps of adopting the existing wavelet threshold denoising and self-adaptive inductance compensation processing to acquired alternating-current time domain waveform data to obtain a pure resistive voltage signal only comprising resistance voltage drop; dividing the processed pure resistive voltage signal and current signal into N (N > 10) sections according to time sequence, and primarily calculating resistance value based on ohm law in each section to obtain multiple groups of sectional resistance data; Extracting pure resistive voltage data URk and current data Ik of the kth segment, and calculating an initial resistance value Rk in each segment k based on ohm's law; taking standard deviation of initial resistance values Rk of all the segments as a weight coefficient sigma k, and carrying out standardization processing on the weight coefficient sigma k to obtain initial weight wk; Based on the initial weight wk, an initial equivalent resistance value RD is calculated by a weighted average method, and the initial equivalent resistance value rd= Σrk×wk.
4. The method for testing and analyzing the time-domain alternating-current equivalent direct-current resistance based on big data according to claim 3, wherein the initial weight wk is subjected to weight updating to obtain wk (b+1), meanwhile equivalent resistance is updated to obtain RD (b+1), and iteration is stopped when the absolute value of RD (b+1) -RD (b) is less than or equal to a preset threshold value, and RD (b+1) is the final equivalent direct-current resistance value.
5. The method for testing and analyzing the time-domain alternating-current equivalent direct-current resistance based on big data according to claim 4, wherein the equivalent direct-current resistance ZR and the final equivalent resistance RD (b+1) are compared and analyzed, the difference between the equivalent direct-current resistance ZR and the final equivalent resistance RD (b+1) is set as an equivalent direct-current resistance deviation, and the equivalent direct-current resistance deviation is judged to obtain an effective signal or an ineffective signal result; When an effective signal is generated, the calculated equivalent direct current resistance ZR, the acquisition time and the equipment operation condition data are stored in a database in a correlated way, and meanwhile, a resistance-condition correlation library is constructed.
6. The method for testing and analyzing time-domain alternating-current equivalent direct-current resistance based on big data according to claim 1, wherein the state analysis process of the dual-model studying and judging device is as follows: extracting characteristic parameters of the power equipment from the database to form a characteristic vector; Inputting the feature vector into a geometric cascading forest algorithm model to obtain class probability vectors of the L-th layer ML geometric forest, carrying out arithmetic average on the collected ML probability vectors according to class dimensions to obtain a final average probability vector, comparing probability values of all classifications (including normal, potential risks and abnormal) in the final average probability vector, and selecting the class with the largest probability as a current final prediction result.
7. The big data-based time-domain alternating-current equivalent direct-current resistance testing and analyzing method according to claim 6, wherein the method is characterized in that in addition to extracting the time sequence characteristics of equivalent resistance values, working condition association characteristics are additionally introduced to form a high-dimensional characteristic matrix; Inputting the high-dimensional feature matrix into a fusion model of a gradient lifting tree and an isolated forest to obtain an output real-time health index and real-time isolation, comparing and analyzing the real-time health index to obtain a normal state result, a potential risk state result and an abnormal state result, and collectively called a state grade; and comparing and analyzing the real-time isolation degree to obtain a potential failure-free result and a potential failure-free result, and collectively called a failure result.
8. The big data based time domain alternating current equivalent direct current resistance test and analysis method according to claim 7, wherein the state grade and the fault result are subjected to interactive analysis, and a final state discrimination result is obtained based on the interactive analysis, wherein the final state discrimination result comprises normal, potential risk and abnormality; And comparing and analyzing the final state judging result with the final predicting result, outputting a final predicting result display if the final state judging result is consistent with the final predicting result, generating a deviation signal if the final state judging result is inconsistent with the final predicting result, immediately responding to the deviation signal, and constructing an abnormal feedback list based on the final state judging result, the final predicting result, the real-time health index, the real-time isolation degree, the fault result, the state grade and the probability value of each classification.

Description

Time domain alternating current equivalent direct current resistance test and analysis method based on big data Technical Field The invention relates to the technical field of power equipment detection and state evaluation, in particular to a time domain alternating current equivalent direct current resistance test and analysis method based on big data. Background The direct current resistance of the power equipment is a core parameter reflecting the health states of key parts such as equipment windings, joints and the like, the numerical variation of the direct current resistance directly relates to fault hidden dangers such as poor contact, winding aging, insulation damage and the like of the equipment, the traditional direct current resistance test needs to cut power to the equipment, the power supply continuity is affected, the test period is long, the efficiency is low, and the requirement of 'state overhaul' of a modern power system is difficult to meet. At present, the AC equivalent DC resistance testing method is mostly based on a single signal processing algorithm, has the problems of weak anti-interference capability and insufficient testing precision, meanwhile, the analysis of test data is mostly dependent on a single model, equipment working condition information and multidimensional characteristics are not fully fused, so that the state research and judgment accuracy is limited, potential faults are difficult to effectively identify, and a solution is provided for the problem. Disclosure of Invention The invention aims to provide a time domain alternating current equivalent direct current resistance test and analysis method based on big data, which realizes high-precision test and accurate identification of health state of the equivalent direct current resistance of power equipment through double-path resistance calculation, multi-dimensional feature fusion, double-mode research and judgment and cross verification. The invention aims at realizing the time domain alternating current equivalent direct current resistance test and analysis method based on big data, which comprises the following steps: The method comprises the steps of firstly, calculating an equivalent direct current resistance through a dual path, namely, collecting sample data to construct a prediction model, and calculating in real time to obtain an equivalent direct current resistance ZR; the validity of the resistor is checked, namely, the deviation between ZR and RD is calculated, and an effective signal is generated when the deviation is within a preset range, otherwise, an ineffective signal is generated and early warning is triggered; step three, constructing characteristics and storage data, namely after effective signals are generated, associating storage resistance, time and working condition data, and simultaneously fusing resistance time sequence characteristics and working condition association characteristics to construct a high-dimension characteristic matrix and a basic characteristic vector; Inputting the basic feature vector into a geometric cascading forest model to obtain a first prediction result, inputting the high-dimensional feature matrix into a gradient lifting tree and isolated forest fusion model, and combining with interactive analysis to obtain a second prediction result; And fifthly, verifying and feeding back the results, namely comparing the two prediction results, outputting the results if the two prediction results are consistent, generating a deviation signal if the two prediction results are inconsistent, and constructing an abnormal feedback list. Preferably, the process of obtaining the equivalent direct current resistor ZR is as follows: s1, collecting alternating-current time domain waveform data of n (n > 5) groups of power equipment as sample data, and extracting two core variables (a predicted variable and a reference variable) in the sample data; carrying out standardization processing on the predicted variable, and carrying out iteration processing on the predicted variable and the reference variable which are subjected to standardization processing through a set least square method to obtain a final coefficient vector beta; S2, acquiring alternating current time domain waveform data of m (m > 0) groups of power equipment in real time, and carrying out standardization processing on the alternating current time domain waveform data of each group to obtain a prediction variable Xnew; s3, calculating to obtain equivalent direct current resistance values of each group based on R=prediction variable Xnew×final coefficient vector beta; And S4, carrying out average value calculation on the equivalent direct current resistance value, and taking the average value calculation result of the equivalent direct current resistance value as a final equivalent direct current resistance ZR. Preferably, the process of obtaining the final equivalent resistance RD is as follows: the method comprises the