CN-121993365-A - Fault early warning method and system for wind farm operation and maintenance equipment

CN121993365ACN 121993365 ACN121993365 ACN 121993365ACN-121993365-A

Abstract

The invention discloses a fault early warning method and a system of wind farm operation and maintenance equipment, which relate to the technical field of fan monitoring and comprise the steps of deploying a multi-mode sensor on a fan to collect data and process the data, extracting data vectors, carrying out vector fusion through a deep branch encoder and outputting fan abnormal probability; the method comprises the steps of constructing a scattering invariant based on data vectors, defining a third-order coupling quantity, synchronously constructing a third-order amplitude through fusion vector decomposition, calculating a third-order residual error by combining the third-order coupling quantity and the third-order amplitude, outputting the third-order residual error, mapping topology according to wind turbine energy transmission, calculating a propagation core of the topology, calculating a topology weight according to the propagation core, outputting an eikonal core factor and shape correction, and finally outputting a wind turbine eikonal risk index, constructing a wind turbine topological graph, defining an initial characteristic of wind turbine nodes, and analyzing a wind turbine abnormal node source through a graph volume integration algorithm. The invention obviously reduces false alarm and missing report, and has high sensitivity and strong robustness under the disturbance of small sample abnormal and complex working condition.

Inventors

WANG PING
SHI MIN
CAI HONGJUN
YANG HUIQIANG
YANG LIBIN
XIAO RUI
ZHANG WENBO
HOU PENG
ZHANG SILIANG
ZHAO HAIFENG
LI MINGDONG
Hu Qingtu
WANG LEI
XU YIWEI
HE XIAODI
HAO YONG
ZHOU HONGREN

Assignees

北京京能电力股份有限公司乌兰察布分公司
内蒙古电力勘测设计院有限责任公司

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (10)

1. A fault early warning method for wind farm operation and maintenance equipment is characterized by comprising the steps of, Deploying a multi-mode sensor on the fan to collect and process data, extracting data vectors, carrying out vector fusion through a depth branch encoder, and outputting fan abnormal probability; Constructing a scattering invariant based on the data vector, defining a third-order coupling amount, synchronously constructing a third-order amplitude through fusion vector decomposition, and calculating a third-order residual error by combining the third-order coupling amount and the third-order amplitude and outputting the third-order residual error; According to the wind turbine energy transfer mapping topology and calculating the propagation kernel of the topology, calculating the topology weight according to the propagation kernel, outputting eikonal core factors and shape correction, and finally outputting a wind turbine eikonal risk index; and constructing a fan topological graph, defining fan node initial characteristics, and analyzing a fan abnormal node source through a graph volume integration algorithm.
2. The method for early warning faults of wind farm operation and maintenance equipment according to claim 1, wherein the steps of deploying the multi-mode sensor on the fan to collect and process data are that vibration sensors, acoustic sensors, visual sensors and environment sensors are installed on the fan to collect vibration, sound, image and environment data of the fan, time synchronization is conducted on the data collected by the multi-mode sensor, a time window is defined to segment the collected data, filtering denoising processing is conducted through a Hamming window and a linear phase FIR low-pass filter, and standardized processing is conducted on the denoised data.
3. The method for early warning faults of wind farm operation and maintenance equipment according to claim 2, wherein the steps of extracting data vectors, carrying out vector fusion through a depth branch encoder and outputting abnormal probability of a fan are that extracting corresponding data vectors by adopting a characteristic engineering method respectively aiming at vibration, sound, images and environmental data after normalization processing; processing the data vector by adopting a branch encoder, and inputting the data vector into each branch encoder to obtain low-dimensional embedding; calculating normalized branch attention weights of data vectors, respectively And obtaining a fusion vector F; outputting fan abnormality probability by inputting fusion vector into logistic regression head 。
4. The method for early warning faults of wind farm operation and maintenance equipment according to claim 3, which is characterized in that scattering invariants are built based on data vectors, third-order coupling quantity is defined, third-order amplitude is built through fusion vector decomposition synchronously, third-order residual errors are calculated by combining the third-order coupling quantity and the third-order amplitude, and the scattering invariants are built based on the sum norm square of the extracted data vectors in Euclidean space; Aligning the cross-modal coupling of the scattering invariant and the data vector to construct a first-order coupling quantity; Calculating the consistency quantity of the three-mode multiplication according to the modal correlation in the first-order coupling quantity, and constructing a second-order coupling quantity; defining four-mode 'cluster' consistency and mapping to six third-order primitives to construct third-order coupling quantity; Constructing a first-order to third-order amplitude by using tensor kernels decomposed by the fusion vector F through a low-rank CP; calculating a third-order residual error by combining the third-order coupling amount and the third-order amplitude, and outputting the third-order residual error; and outputting a fusion vector, an anomaly probability, a third-order coupling amount, a third-order amplitude and a third-order residual error for each time window m.
5. The method for early warning faults of wind farm operation and maintenance equipment according to claim 4, which is characterized in that the method comprises the steps of mapping topology according to wind turbine energy transmission and calculating transmission cores of the topology, calculating topology weights according to the transmission cores and outputting eikonal core factors and shape correction, wherein finally outputting wind turbine eikonal risk indexes means mapping into four topology routes according to wind turbine energy transmission routes; for fan components in each topological route, counting average spectrum peaks of sensors of the components in a health period, and taking a mode index with the maximum average spectrum peak as And calculates the green's function of the component in the topology propagation core ; Respectively calculating topology propagation kernels for four types of topology routes based on green functions of the fan components; Defining a topology average core strength according to the propagation cores of each topology; Generating a step coupling score based on the third-order coupling quantity; Calculating a step weight by combining the step coupling score and normalizing; calculation of full-order expansion factor for current time window to obtain eikonal core factor Shape correction ; Calculation of the eikonal risk index of a blower in combination with the eikonal core factor and shape correction 。
6. A fault pre-warning method for wind power plant operation and maintenance equipment is characterized in that a fan topological graph is built and initial characteristics of fan nodes are defined, fan abnormal node sources are analyzed through a graph volume integration algorithm, the fan component topological graph is built according to components of a fan, connecting edges are built according to relations among nodes and undirected processing is carried out in GCN, and node initial characteristics are given to each node ; Adopting two layers of GCNs to combine with residual errors to forward propagate node characteristics; and finally outputting calculated node suspicion degree according to the GCN, and taking the node with the highest suspicion degree as an abnormal node source for early warning.
7. The method for fault pre-warning of wind farm operation and maintenance equipment according to claim 6, wherein after the abnormal node source of the fan is obtained, abnormal node marking is carried out in a three-dimensional digital model of the fan and highlighting is carried out.
8. A fault early warning system of wind farm operation and maintenance equipment is characterized by comprising the following steps of, The anomaly analysis module is used for deploying a multi-mode sensor on the fan to collect and process data, extracting data vectors, carrying out vector fusion through the deep branch encoder and outputting fan anomaly probability; the structure mapping module is used for constructing a scattering invariant based on the data vector, defining a third-order coupling quantity, synchronously constructing a third-order amplitude through fusion vector decomposition, and calculating a third-order residual error by combining the third-order coupling quantity and the third-order amplitude and outputting the third-order residual error; The structural interference analysis module is used for mapping topology according to wind turbine energy transfer and calculating a transmission core of the topology, calculating a topology weight according to the transmission core, outputting eikonal core factors and shape correction, and finally outputting a wind turbine eikonal risk index; The topology analysis module is used for constructing a fan topology graph and defining the initial characteristics of fan nodes, and analyzing the sources of abnormal fan nodes through a graph volume integration algorithm.
9. The computer equipment comprises a memory and a processor, wherein the memory stores a computer program, and the computer program is characterized in that the processor realizes the steps of the fault early warning method of the wind farm operation and maintenance equipment according to any one of claims 1-7 when executing the computer program.
10. A computer readable storage medium, on which a computer program is stored, is characterized in that the computer program, when being executed by a processor, implements the steps of the fault pre-warning method for wind farm operation and maintenance equipment according to any one of claims 1 to 7.

Description

Fault early warning method and system for wind farm operation and maintenance equipment Technical Field The invention relates to the technical field of fan monitoring, in particular to a fault early warning method and system for wind farm operation and maintenance equipment. Background The wind power plant operation and maintenance (O & M) state monitoring and fault early warning technology is subjected to three evolutions from single sensor monitoring to multi-mode sensing, threshold judgment to intelligent diagnosis and off-line analysis to edge calculation. The early stage uses SCADA and accelerometer as cores, realizes coarse granularity early warning by an empirical threshold method which depends on characteristics such as root mean square, kurtosis, envelope spectrum and gear meshing side band, and then adopts a data driving method represented by HMM, SVM, RF and the like, and then adopts a deep learning method of inputting a vibration spectrogram/sound spectrogram into CNN, inputting a long time sequence working condition into LSTM or CRNN, realizes automatic identification of typical faults such as bearing pitting, tooth surface abrasion, blade icing and the like, introduces multi-modal fusion (vibration/acoustics/vision/strain/environment), digital twinning and edge deployment in recent years, explores to couple a device physical model with a data model for improving timeliness and interpretability, and simultaneously, tries to characterize a fault propagation track from the angle of 'component-connection-energy flow' in the aspect of complex system topological modeling. In general, the industry has moved from 'punctiform threshold monitoring' to 'system-level space-time modeling', but the prior art still has the defects, the common early/late fusion is mainly based on empirical weighting or black box splicing, and the lack of verifiable cross-mode consistency measurement is easy to generate false alarms in strong wind and turbulence scenes, so that the requirement of high-precision fan operation monitoring cannot be met. 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 fault early warning method and system for wind farm operation and maintenance equipment, and solves the problems that the prior art lacks verifiable cross-mode consistency measurement, false alarms are easy to generate in strong wind and turbulence scenes, and the requirements of high-precision fan operation monitoring cannot be met. In order to solve the technical problems, the invention provides the following technical scheme: in a first aspect, the invention provides a fault early warning method for wind farm operation and maintenance equipment, which comprises the steps of, Deploying a multi-mode sensor on the fan to collect and process data, extracting data vectors, carrying out vector fusion through a depth branch encoder, and outputting fan abnormal probability; Constructing a scattering invariant based on the data vector, defining a third-order coupling amount, synchronously constructing a third-order amplitude through fusion vector decomposition, and calculating a third-order residual error by combining the third-order coupling amount and the third-order amplitude and outputting the third-order residual error; According to the wind turbine energy transfer mapping topology and calculating the propagation kernel of the topology, calculating the topology weight according to the propagation kernel, outputting eikonal core factors and shape correction, and finally outputting a wind turbine eikonal risk index; and constructing a fan topological graph, defining fan node initial characteristics, and analyzing a fan abnormal node source through a graph volume integration algorithm. The method for early warning faults of wind farm operation and maintenance equipment comprises the steps of arranging a multi-mode sensor on a fan to collect data and processing the data, wherein vibration sensors, acoustic sensors, visual sensors and environment sensors are arranged on the fan to collect vibration, sound, images and environment data of the fan, time synchronization is carried out on the data collected by the multi-mode sensor, a time window is defined to segment the collected data, filtering denoising processing is carried out by combining a Hamming window with a linear phase FIR low-pass filter, and standardized processing is carried out on the denoised data. The method for early warning faults of wind farm operation and maintenance equipment comprises the steps that extracted data vectors are subjected to vector fusion through a depth branch encoder and fan abnormal probability is output, and corresponding data vectors are extracted by adopting a characteristic engineering method respectively aiming at vibration, sound, images and environmental data after normalization processing; processing the data vector by adopting a b