CN-121980401-A - Power equipment multi-source fault detection method based on data fusion

CN121980401ACN 121980401 ACN121980401 ACN 121980401ACN-121980401-A

Abstract

The invention discloses a multi-source fault detection method of power equipment based on data fusion, which relates to the technical field of power equipment faults and comprises the steps of synchronously collecting ultrahigh frequency, high frequency, ultrasonic, transient ground voltage and infrared/voiceprint data under a unified time reference to generate unified timestamp frame streams, checking and splicing the frame streams, fixing window slices, normalizing noise bottom quantity after frequency band pretreatment to obtain abnormal degree, determining event windows and outputting event data packets, extracting a minimum feature set to calculate channel quality coefficients, classifying and screening out suspected fault events, randomly outputting fault type probability and combining channel quality to form evidence packets, calculating relevant consistency items and combining with association rules to obtain rule consistency items, fusing to form consistency indexes, executing D-S combination to output fused fault distribution, conflict degree and fusion uncertainty after evidence is discounted, and calculating growth trend quantity, thereby realizing pre-filtering of non-fault events and probability output of fault types.

Inventors

YAO LI
DAI CHEN
JIANG DAPENG
ZHANG HAO
WANG LEI

Assignees

中广核(安徽)新能源投资有限公司

Dates

Publication Date: 20260505
Application Date: 20251229

Claims (10)

1. A multi-source fault detection method for power equipment based on data fusion is characterized by comprising the following steps of, Establishing a detection task, synchronously collecting monitoring data under the same time reference, and generating a multi-source data frame stream with uniform time stamps; Checking and splicing based on the multi-source data frame stream, slicing according to a fixed window to form candidate windows, preprocessing the frequency bands of all channels, calculating window anomaly degree of all channels through noise floor normalization, determining event windows, and outputting event data packets; Extracting a minimum feature set from an event data packet, calculating a channel quality coefficient, gating an event window by using a two-class model, screening out suspected fault events, outputting probability distribution for the suspected fault events by using a random forest classifier, and combining the probability distribution with the channel quality coefficient to obtain an evidence packet; calculating a relevant consistency item through a preprocessing waveform of an event window, calculating a rule consistency item based on a preset association rule set and an evidence packet, fusing the relevant consistency item and the rule consistency item to form a consistency index, discounting the evidence of each channel by the consistency index and the channel quality coefficient, and then executing D-S combination to output fused fault distribution, conflict degree and fused uncertainty; and judging whether to trigger retest or not by fusing fault distribution, conflict degree and fusion uncertainty, calculating the growth trend quantity and writing in a detection result record.
2. The method for detecting the multi-source fault of the power equipment based on the data fusion of claim 1, wherein the method is characterized in that the detection task is established and the monitoring data are synchronously collected under the same time reference to generate a multi-source data frame stream with uniform time stamps, Establishing a detection task, synchronously collecting monitoring data under the same time reference, outputting a primary synchronous trigger pulse to a partial discharge related channel, and completing alignment by adopting a secondary trigger calibration method; Continuously sampling the ultra-high frequency, ultrasonic and transient voltage channels, writing a master clock time stamp into each sampling point, writing a master clock time stamp corresponding to each image recording frame into each infrared channel, writing an image frame head into each image recording frame, and collecting array data with fixed frame length into a voiceprint array channel to obtain a multi-source data frame stream.
3. The method for detecting the multi-source fault of the power equipment based on the data fusion as claimed in claim 2, wherein the steps of checking, splicing and slicing according to fixed window to form candidate windows are carried out on the multi-source data frame flow, the frequency band preprocessing is carried out on each channel, Performing consistency check on each frame in the multi-source data frame stream through CRC, and writing a missing mark if the adjacent two frames of the same channel are discontinuous or not increased; The window length and the sliding step length are fixed, a candidate window sequence is generated by taking a uniform time stamp as a reference, the deletion proportion of the candidate window is calculated through the deletion marking, and when the deletion proportion is larger than a deletion threshold value, the candidate window is marked as too many deletions; and performing frequency band preprocessing on the partial discharge related channels in a candidate window, performing convolution operation on the original fragments in the candidate window to obtain band-pass output, performing saturation detection on each partial discharge related channel, and setting a sample ratio of a sampling value reaching the upper limit of the range as a saturation mark if the sample ratio exceeds a saturation threshold.
4. The method for detecting the multi-source fault of the power equipment based on data fusion as claimed in claim 3, wherein the step of calculating the window anomaly degree of each channel through noise floor normalization, determining an event window and outputting an event data packet comprises the following specific steps of, Maintaining a noise floor quantity for each partial discharge related channel, taking window energy median of a fixed number of windows before the task starts, and calculating the anomaly degree for each candidate window and each partial discharge related channel; And maintaining an abnormality degree queue for each partial discharge related channel, marking the channel as abnormal and effective in the current window when the abnormality degree of the channel in the candidate window is larger than a dynamic trigger threshold and the candidate windows with excessive and saturated marks are not deleted, judging the candidate windows as event windows, and packaging the event windows into event data packets.
5. The method for detecting multi-source fault of power equipment based on data fusion as claimed in claim 4, wherein the steps of extracting the minimum feature set from the event data packet and calculating the channel quality coefficient are as follows, Performing pulse detection on the event window band-pass output, counting the number of pulse segments exceeding a pulse detection threshold in the event window as a pulse counting feature, taking the ratio of the maximum phase bucket count to the total count as an aggregation degree in a channel window, acquiring a phase aggregation feature, and taking the ratio of the energy detected as the pulse segments in the window to the total energy of the window as a pulse energy occupation feature; And splicing the normalized energy characteristic, the pulse counting characteristic, the phase aggregation characteristic and the pulse energy occupation characteristic into a minimum characteristic set in sequence, and calculating a channel quality coefficient in an event window for each partial discharge related channel.
6. The method for detecting the multi-source fault of the power equipment based on the data fusion as claimed in claim 5, wherein the method is characterized in that the method comprises the steps of gating an event window by using a classification model, screening out suspected fault events, outputting probability distribution to the suspected fault events by using a random forest classifier, combining the probability distribution with a channel quality coefficient to obtain an evidence packet, Inputting the minimum feature set into the two classification models, and marking the current event window as a suspected fault event when the two classification models are output as suspected faults; based on suspected fault events, multi-classification reasoning is executed on each partial discharge related channel based on a random forest classifier respectively, probability distribution is obtained, and the probability distribution is combined with channel quality coefficients to obtain evidence packets.
7. The method for detecting multi-source faults of power equipment based on data fusion as claimed in claim 6, wherein the preprocessing waveform through the event window calculates a relevant consistency term, and the rule consistency term is calculated based on a preset association rule set and an evidence package, specifically comprising the following steps of, Calculating a correlation consistency term for the preprocessed waveforms in the same event window; and setting an association rule set, judging the minimum feature set rule by rule based on the evidence packet, respectively counting the number of the supporting rule hits and the number of the conflict rule hits, and calculating a rule consistency item.
8. The method for detecting multi-source faults of power equipment based on data fusion as claimed in claim 7, wherein the method is characterized in that the related consistency items and the rule consistency items are fused to form consistency indexes, D-S combination is executed after discounting the evidence of each channel by the consistency indexes and the channel quality coefficients, and fusion fault distribution, conflict degree and fusion uncertainty are output, The relevant consistency items and the rule consistency items are fused to form consistency indexes, and the consistency indexes and the channel quality coefficients are used for jointly discounting the evidence of each channel to obtain quality distribution; D-S combination is carried out on the evidence after discount of each channel, the conflict degree is calculated, and the conflict degree obtained by the last combination is used as the total conflict degree of the event window; and extracting and normalizing the quality distribution of each fault type to obtain a fusion fault distribution, and simultaneously calculating the fusion uncertainty.
9. The method for detecting the multi-source fault of the power equipment based on data fusion as claimed in claim 8, wherein the step of judging whether to trigger retesting by fusion of fault distribution, conflict degree and fusion uncertainty comprises the following steps of, Executing retest triggering judgment on each suspected fault event window, and judging that the reliability of the current fusion result is insufficient when the fusion uncertainty is higher than a fusion uncertainty judgment threshold value, or the total conflict degree is higher than a conflict judgment threshold value, or the consistency index is lower than a consistency judgment threshold value, so that retest is needed; when the complex is determined to be needed, selecting a unique complex measurement action, executing the complex measurement action, calculating a unit energy consumption fusion uncertainty reduction index, and selecting the complex measurement action with the largest unit energy consumption fusion uncertainty reduction index as the unique complex measurement action.
10. The method for detecting the multi-source fault of the power equipment based on the data fusion as claimed in claim 9, wherein the steps of calculating the growth trend amount and writing the detection result record are as follows, And after the retest action is executed, receiving the retest data correspondingly generated, recalculating the fusion uncertainty, the overall conflict degree and the consistency index, calculating the fault growth trend when the fusion uncertainty, the overall conflict degree and the consistency index are lower than the respective judgment threshold, and writing the fault growth trend and the fusion fault distribution into the detection result record.

Description

Power equipment multi-source fault detection method based on data fusion Technical Field The invention relates to the technical field of power equipment faults, in particular to a power equipment multi-source fault detection method based on data fusion. Background The power equipment fault detection technology is continuously evolved along with the development of on-line monitoring and digital operation and maintenance, a multi-parameter collaborative acquisition system is gradually formed around the local discharge related detection means of a high-voltage switch cabinet, a transformer, cable accessories and the like, meanwhile, the improvement of the calculation capability of a mobile terminal and an edge promotes the on-site deployment of event windowing processing, feature extraction and intelligent recognition algorithms, and related research is developed from single-channel threshold judgment to multi-source fusion, probabilistic diagnosis and comprehensive evaluation oriented to fault prediction and health management. However, the existing method still has two limitations that firstly, in the engineering implementation of multi-source data frame flow, common practice only carries out simple data receiving and buffering, and lacks frame-by-frame consistency check and frame sequence continuity check, so that abnormal frames, lost frames and disordered sequences caused by transmission disturbance or link jitter are difficult to identify in time and form traceable marks, further, hidden gaps exist in subsequent window slice and channel alignment to influence fault detection stability, secondly, local discharge signals are easy to generate out-of-band noise superposition and front end saturation distortion under the conditions of strong interference and limited range, and if the existing method does not carry out clear frequency band pretreatment on local discharge related channels and carries out threshold judgment isolation on saturation duty ratio, the distorted high-amplitude fragments are often mistaken as effective abnormal features. 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 multi-source fault detection method of power equipment based on data fusion, which solves the problems of characteristic distortion caused by the lack of consistency check and traceable marks for abnormal data frames, frame loss and disorder and the lack of effective isolation for out-of-band interference and saturation distortion of partial discharge channels in the prior art. In order to solve the technical problems, the invention provides the following technical scheme: The invention provides a multi-source fault detection method of power equipment based on data fusion, which comprises the steps of establishing a detection task, synchronously collecting monitoring data under the same time reference, and generating a multi-source data frame stream with uniform time stamps; Checking and splicing based on the multi-source data frame stream, slicing according to a fixed window to form candidate windows, preprocessing the frequency bands of all channels, calculating window anomaly degree of all channels through noise floor normalization, determining event windows, and outputting event data packets; Extracting a minimum feature set from an event data packet, calculating a channel quality coefficient, gating an event window by using a two-class model, screening out suspected fault events, outputting probability distribution for the suspected fault events by using a random forest classifier, and combining the probability distribution with the channel quality coefficient to obtain an evidence packet; calculating a relevant consistency item through a preprocessing waveform of an event window, calculating a rule consistency item based on a preset association rule set and an evidence packet, fusing the relevant consistency item and the rule consistency item to form a consistency index, discounting the evidence of each channel by the consistency index and the channel quality coefficient, and then executing D-S combination to output fused fault distribution, conflict degree and fused uncertainty; and judging whether to trigger retest or not by fusing fault distribution, conflict degree and fusion uncertainty, calculating the growth trend quantity and writing in a detection result record. The method for detecting the multi-source faults of the power equipment based on data fusion is used as a preferable scheme, wherein the method establishes a detection task and synchronously collects monitoring data under the same time reference to generate a multi-source data frame stream with uniform time stamps, and comprises the following specific steps, Establishing a detection task, synchronously collecting monitoring data under the same time reference, outputting a primary synchronous trigger pulse to a partial discharge related