CN-122020109-A - Converter valve recording data extraction method based on self-adaptive filtering and data driving

CN122020109ACN 122020109 ACN122020109 ACN 122020109ACN-122020109-A

Abstract

A converter valve recording data extraction method based on self-adaptive filtering and data driving belongs to the technical field of converter valve fault identification and comprises the following steps of extracting and processing recording data of converter valves under different working conditions, wherein the recording data comprise core variable parameters under each converter valve fault type, constructing a core variable parameter sensitivity analysis calculation model based on the converter valve fault type, extracting core variable parameter sensitivity under each converter valve fault type, sorting according to the sensitivity of the core variable parameters, constructing a 3D-CNN three-dimensional convolutional neural network, constructing corresponding three-dimensional time-frequency representation through a core variable parameter sensitivity sorting result as network input, introducing an self-attention module to highlight key features, and finally extracting high-dimensional feature vectors which can accurately represent the fault type from a network intermediate layer. Through the improvement of an algorithm, the accurate and rapid extraction of the fault recording data characteristics of the converter valve is realized.

Inventors

ZHANG XIAOFEI
XU JINGZHENG
LIU YANG
CHEN YUFEI
LIU XIANHUI
LI TAO
Teng Mengfeng
WANG YISHUO
WANG JIAN
ZHAO CHUANGANG
LI XIANJUN
LI PENG
MENG LINGJUN
ZHANG ZONGBAO
TANG JIANLI
LIU LUQI
WANG SHUYUAN
CHEN FANGDONG
JIN HAIWANG
LI JINBO
SONG JIANLIANG
Tian Kaizhe
YU WENBO

Assignees

国网冀北电力有限公司超高压分公司
国家电网有限公司
山东山大电力技术股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251205

Claims (10)

1. The converter valve recording data extraction method based on the self-adaptive filtering and the data driving is characterized by comprising the following steps of: Step S10, extracting and processing recording data of converter valves under different working conditions, wherein the recording data comprise core variable parameters under each converter valve fault type, and the core variable parameters comprise converter valve equipment parameters and power grid interference intensity; step S20, constructing a core variable parameter sensitivity analysis calculation model based on the fault types of the converter valves, extracting the core variable parameter sensitivity degree under each fault type of the converter valves, and sequencing according to the sensitivity degree of the core variable parameter; And S30, constructing a 3D-CNN three-dimensional convolutional neural network, constructing a corresponding three-dimensional time-frequency representation through a core variable parameter sensitivity degree sequencing result as network input, introducing a self-attention module to highlight key features, and finally extracting a high-dimensional feature vector which can accurately represent the fault type from a network middle layer.
2. The method for extracting the recording data of the converter valve according to claim 1, wherein the recording data of the converter valve in step S10 is further subjected to data cleaning, and the analysis method of the data cleaning process is as follows: Firstly, eliminating extreme outliers in the wave recording data by adopting 3 sigma-Graves fusion criteria; Secondly, performing value-missing complement processing on the recording data by adopting a time sequence empty window linear-spline mixed interpolation complement algorithm.
3. The method for extracting data from a converter valve recording data according to claim 1, wherein, The fault type of the converter valve comprises sub-module faults, the fault type mark is 1, the synchronous faults among the modules, the fault type mark is 2, and the fault types of other types are 3; The collection of the fault types of the converter valve is F type = c, c= {1, 2, 3}, and c represents a fault type identifier; The converter valve equipment parameters comprise actual measurement voltage U value at two ends of a converter valve, current I arm flowing through a bridge arm of the converter valve, equivalent submodule capacitance value C eq and submodule switching state S mod ; The power grid interference intensity comprises a composite index F dip formed by the total harmonic distortion rate THD u of alternating-current side voltage and the voltage sag depth and duration.
4. The method for extracting data from converter valve recording data according to claim 1, wherein the process of constructing the core variable parameter sensitivity analysis calculation model in step S20 includes: Step 201, constructing a feature set V c of core variable parameters of each converter valve fault type, wherein each parameter is a vector, the length is N, V c ＝{V 1 ,V 2 ,...,V K }＝{U value ,I arm ,C eq ,S mod ,THD u ,F dip }, Wherein K represents the number of core variable parameters; step 202, constructing a random forest-based fault identification model And a fault recognition accuracy model, Wherein I is an indication function, 1 is taken when the value of I meets the condition, and 0 is taken otherwise; step 203, eliminating the fault identification precision model after the k-th core variable parameter participates in, Wherein the method comprises the steps of V c,k represents the feature set from which the kth core variable parameter is removed; in step 204, the sensitivity degree formula of the kth variable to the fault type identification precision model, Importance k ＝Acc-Acc k , The larger the value of Importance k , the greater the sensitivity of the kth core variable parameter to the fault type identification; Step 205, outputting the core variable parameter sensitivity degree order V' of each fault type, V′ c ＝{V (1) ,V (2) ,...,V (K) }, Wherein V c ' represents the sensitivity degree order of all the core variable parameters in the fault type c, V (1) represents the core variable parameter with the highest sensitivity degree order, and V (K) represents the core variable parameter with the lowest sensitivity degree order; step 206, screening out the top three core variable parameters of the sensitivity level of each fault type as a key variable set, V′ selected ＝{V (1) ,V (2) ,V (3) }。
5. The method for extracting the recording data of the converter valve according to claim 1, wherein the step S30 of constructing the corresponding three-dimensional time-frequency process from the core variable parameter sensitivity degree ordering result includes: Firstly, dividing each core variable parameter into D continuous fragments according to a fixed time window length T, wherein each fragment comprises L sampling points, carrying out short-time Fourier transform on signal segments in each time window, and the mathematical expression is that, F=(L/2)+1, f={0.1...F-1}, d={0.1...D-1}, a={0.1...L-1}, Wherein x k (a) represents the amplitude of the core variable parameter V k at the a-th sampling point, ω (·) is a window function, d is a time window index, F is a frequency index, H is a window shift step length, and F is the number of frequency points; secondly, obtaining a time-frequency matrix corresponding to the core variable parameter V k through the transformation Matrix element X k (d, f) represents the spectral amplitude at time window d, frequency point f; finally, stacking all time-frequency matrixes of the selected core variable parameters along the channel dimension to construct three-dimensional time-frequency input, C=|V′ selected |; And taking the three-dimensional time frequency as the input of the 3D-CNN three-dimensional convolution neural network, sliding the convolution kernel W on the time and frequency dimensions, and generating a feature map Z through convolution operation.
6. The method according to claim 1, wherein in step 30, a self-attention module is introduced after the convolution layer processing, and calculates an attention weight for each feature vector in the feature map Z to reflect the importance of the attention weight on the fault classification.
7. The method for extracting the converter valve recording data according to claim 1, wherein in the step 30, an expression capable of accurately characterizing the high-dimensional feature vector v c corresponding to the fault type is extracted from the network intermediate layer: Where Z ' d,f,c is the value of the c-th fault type of the signature Z ' at position (d, f), Z ' represents the signature output from the attention module.
8. An electronic device comprising a memory, a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1-7 when executing the computer program.
9. A non-transitory computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method of any of claims 1-7.
10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any of claims 1-7.

Description

Converter valve recording data extraction method based on self-adaptive filtering and data driving Technical Field The invention belongs to the technical field of converter valve fault identification, and particularly relates to a converter valve recording data extraction method based on self-adaptive filtering and data driving. Background The effective processing and the fault accurate identification of the converter valve recording data are key for guaranteeing the reliable operation of the converter valve. The current converter valve equipment recording data of different manufacturers are various in format, and lack of a unified reading and analyzing frame leads to difficulty in compatible processing of different equipment data. Under the circumstance, the problem of low data preprocessing efficiency is more remarkable, meanwhile, recording data is easily influenced by power grid harmonic waves and electromagnetic interference, and the traditional fixed parameter filtering algorithm cannot dynamically adapt to noise change. In addition, the fault identification is based on the fact that manually extracted experience features are fully related to the fault types and the intrinsic rules of recording data fluctuation features. The problems directly lead to the early warning lag of the converter valve fault and high misjudgment rate, and when serious, the valve group is stopped, the power of the power grid is lacked, and the power supply reliability is damaged. Thus, research on a method for extracting the recording data characteristics of the converter valve is needed. Disclosure of Invention The invention aims to solve the technical problem of providing a converter valve recording data extraction method based on self-adaptive filtering and data driving, and the accurate and rapid extraction of the fault recording data characteristics of the converter valve is realized through the improvement of an algorithm. The technical scheme includes that the method comprises the steps of S10, extracting and processing recording data of a converter valve under different working conditions, wherein the recording data comprise core variable parameters under each converter valve fault type, the core variable parameters comprise converter valve equipment parameters and power grid interference intensity, S20, constructing a core variable parameter sensitivity analysis calculation model based on the converter valve fault type, extracting core variable parameter sensitivity degree under each converter valve fault type, and then sequencing according to the sensitivity degree of the core variable parameters, S30, constructing a 3D-CNN three-dimensional convolutional neural network, constructing corresponding three-dimensional time-frequency representation through the sequencing result of the core variable parameter sensitivity degree as network input, and introducing an automatic attention module to highlight key characteristics, and finally extracting high-dimensional feature vectors which can accurately represent the fault type from a network interlayer. The invention has the beneficial effects that wave recording data of the converter valve under different working conditions are sequentially extracted and processed, the sensitivity of core variables to fault types is analyzed, the priority order of influencing factors is carried out, automatic feature learning and high-dimensional feature vector extraction are carried out through a three-dimensional time-frequency convolution network, a unified data processing frame and multi-dimensional feature expression are provided for realizing the accurate identification and rapid diagnosis of the fault of the converter valve, so that the fault early warning is more timely, the identification precision is obviously improved, the limitation of the traditional method in the treatment of complex interference and dynamic noise is effectively overcome, the characteristics of firm theoretical basis and flexible technical application are achieved, the operation reliability of converter valve equipment is improved, the maintenance cost is reduced, and the safety, stability and economic dispatching of a power grid are ensured. Drawings Fig. 1 is a flow chart of the present invention. Detailed Description In order to make the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Referring to fig. 1, the invention provides a converter valve recording data extraction method based on self-adaptive filtering and data driving, which comprises the following steps of S10, extracting and processing recording data of a converter valve under different working conditions, wherein the recording data comprises core variable parameters under each converter valve f