CN-121976919-A - Fault monitoring method and system for wind turbine generator

CN121976919ACN 121976919 ACN121976919 ACN 121976919ACN-121976919-A

Abstract

The invention relates to a fault monitoring method and system of a wind turbine, wherein mechanical sensor data, environment parameter data and running state data in the wind turbine are collected to be used as an initial data set, data in the initial data set is preprocessed by adopting a data normalization and parameter online updating method to obtain a target data set, dynamic hypergraph modeling is carried out on the running data of the wind turbine according to the target data set, data relation and space-time change are captured, a three-layer hypergraph convolution mode is adopted to construct the space-time hypergraph convolution, feature aggregation is achieved, abnormal detection is carried out by adopting a reconstruction error calculation mode and a mode of classifying and aggregating abnormal features of the wind turbine, and an alarm threshold value of the wind turbine is dynamically adjusted based on real-time running conditions of the wind turbine. According to the method, the vertex set containing multiple types of factors and the space-time correlation superside generation module are constructed, so that the wind turbine is inspected and alarmed in real time, and the effects of rapidly positioning the fault condition of the wind turbine and alarming are achieved.

Inventors

GUO RENHONG
LI JIANPING
Wu guangke
LUO CHENGDE
Fu Meixiao

Assignees

广东粤电珠海海上风电有限公司
西安热工研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20260113

Claims (10)

1. The fault monitoring method for the wind turbine generator is characterized by comprising the following steps of: s1, collecting mechanical sensor data, environmental parameter data and running state data in a wind turbine generator as an initial data set; S2, preprocessing data in an initial data set by adopting a data normalization and parameter online updating method to obtain a target data set; s3, carrying out dynamic hypergraph modeling on the operation data of the wind turbine generator according to the target data set, and capturing data relation and space-time variation; s4, constructing space-time hypergraph convolution by adopting a three-layer hypergraph convolution mode, and realizing feature aggregation; s5, performing anomaly detection by adopting a reconstruction error calculation mode and a mode of classifying and aggregating unit anomaly characteristics; s6, dynamically adjusting an alarm threshold value of the wind turbine generator based on the real-time operation condition of the wind turbine generator.
2. The fault monitoring method for the wind turbine generator according to claim 1, wherein in S1, collecting mechanical sensor data, environment parameter data and running state data in the wind turbine generator as an initial data set comprises capturing and triggering data acquisition actions on a microsecond time scale by utilizing high-speed parallel processing capacity based on a hardware trigger synchronous circuit of a programmable gate array, matching with a Beidou dual-mode high-precision clock timing module, realizing nanosecond timing precision by a built-in atomic clock level time reference, realizing real-time caching and dynamic rearrangement by a hardware pipeline mechanism, ensuring that mechanical, environment and running three-source data are aligned on a time sequence, and further taking the obtained data as the initial data set.
3. The method for monitoring faults of a wind turbine generator according to claim 2, wherein in S2, preprocessing data in an initial data set by adopting a method of data normalization and parameter online updating, and obtaining a target data set includes: eliminating dimension differences among features of the initial data set by adopting a data normalization mode; normalization maps data to the same scale uniformly, and the dynamic normalization formula is as follows: Wherein, the As a result of the normalization of the characteristic values, For raw input feature values, i.e. raw sensor data from the initial dataset, Is the mean value in the K-th case, The standard deviation in the K-th updated case; The method for adjusting the mean value and standard deviation parameters in the normalization process by adopting the online parameter updating mode comprises the following steps: the self-adaptive capacity of the wind turbine generator to different working conditions is improved by adopting an online parameter updating mode, and an online parameter updating formula is as follows: Wherein, the The updated mean value under the K-th working condition, The mean value under the K-th working condition before updating, Calculating the average value of the current data; Wherein, the The standard deviation under the K-th working condition after updating, Standard deviation under the K-th working condition before updating, The standard deviation calculated from the current data.
4. The method for monitoring faults of a wind turbine according to claim 1, wherein in S3, performing dynamic hypergraph modeling on operation data of the wind turbine according to a target data set, capturing a data relationship and a space-time change includes: Step S31, a top point set containing operation data type factors, equipment physical characteristic type factors, environment condition type factors and historical fault related factors is established; step S32, based on the dynamic hypergraph, constructing a space-time correlation hyperedge generation module, and describing the depth of the running state of the wind turbine; Step S33, introducing a sliding time window mechanism based on the space-time correlation factor, and dynamically adjusting the over-edge weight; And step S34, capturing abnormal data by adopting a hypergraph convolution learning and dynamic hypergraph updating mode.
5. The method for monitoring faults of a wind turbine generator according to claim 4, wherein an initial weight is given to each superside based on a space-time correlation factor, similarity is calculated on a nonlinear relation by using a Gaussian kernel function, hidden correlations among nodes are quantized, and the Gaussian kernel function has the following specific formula: Wherein, the As a kernel function, x and y represent two data points in the input space, Is an exponential operation based on a natural constant e; as the square of the distance between data points x and y, Is the bandwidth of the kernel function.
6. The fault monitoring method of a wind turbine generator according to claim 1, wherein in S4, a space-time hypergraph convolution is constructed by adopting a three-layer hypergraph convolution method, and the feature aggregation comprises: and the three-layer hypergraph convolution architecture is adopted for feature aggregation, so that the efficient feature extraction of the multi-source heterogeneous data is realized, and the calculation formula is as follows: Wherein, the Representing characteristic splicing operation; Is a static superside incidence matrix; Is a weight matrix which dynamically changes with time; in order to activate the function, In the form of a degree matrix, Is a matrix of learnable parameters of the first layer.
7. The method for monitoring faults of a wind turbine generator according to claim 1, wherein in S5, performing anomaly detection by adopting a reconstruction error calculation and a method of classifying and aggregating abnormal characteristics of the wind turbine generator comprises: based on a complex topological structure of the wind turbine, the node characteristics are learned and encoded, the reconstruction characteristics are generated, the reconstruction error of each node is calculated, and the calculation formula of the node-level reconstruction error is as follows: Wherein, the Is the first The error metric values of the individual samples, Is the first The actual values of the three-dimensional matrix H of the individual samples, Is the first An estimate of the matrix H of samples.
8. A fault monitoring system for a wind turbine, comprising: The data acquisition unit is used for acquiring mechanical sensor data, environment parameter data and running state data in the wind turbine generator set as an initial data set; the data preprocessing unit is used for preprocessing data in the initial data set by adopting a data normalization and parameter online updating method to obtain a target data set; the dynamic hypergraph modeling unit is used for carrying out dynamic hypergraph modeling on the operation data of the wind turbine generator according to the target data set and capturing the data relationship and the space-time change; The space-time hypergraph convolution construction unit adopts a three-layer hypergraph convolution mode to construct space-time hypergraph convolution so as to realize feature aggregation; The abnormality detection unit is used for carrying out abnormality detection in a mode of calculating reconstruction errors and classifying and aggregating unit abnormality characteristics; and the dynamic adjustment unit is used for dynamically adjusting the alarm threshold value of the wind turbine generator based on the real-time operation condition of the wind turbine generator.
9. The network side service end is characterized by comprising at least one processor and a memory in communication connection with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the fault monitoring method of the wind turbine generator set according to any one of claims 1 to 7.
10. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements a method for fault monitoring of a wind turbine according to any of claims 1 to 7.

Description

Fault monitoring method and system for wind turbine generator Technical Field The invention relates to the technical field of data processing, in particular to a fault monitoring method and system of a wind turbine generator. Background The wind turbine generator is a large-scale power generation device standing on land or at sea, captures the kinetic energy of wind through three giant blades and drives a main shaft to rotate, and then transmits mechanical energy to a generator through a gear box, and finally, the mechanical energy is integrated into a power grid in an electric energy form to provide clean renewable energy. At present, fault monitoring of a wind turbine generator generally adopts the steps of installing vibration, temperature, rotating speed and power sensors at key positions of a cabin, a gear box, a generator and the like, summarizing data to a monitoring background through a PLC or an edge computing gateway, comparing according to a fixed threshold value, spectrum energy, a statistical index or a simplified physical model, and triggering an audible and visual alarm if a certain index is out of limit. However, in the existing wind turbine generator system fault monitoring method, in the actual use process, the phase dislocation can be caused by the asynchronous collected multi-source data clocks, the fixed threshold value which is initially set cannot adapt to the real-time change of wind speed and load, the alarm information only prompts abnormality and cannot directly lock specific components such as blades, bearings or converters, operation and maintenance personnel still need to check layer by layer, and the alarm condition is not classified, so that the false alarm of part of component faults can be caused. Disclosure of Invention Based on the problems in the prior art, the invention aims to provide a fault monitoring method and system for a wind turbine generator. In order to achieve the above purpose, the present invention adopts the following technical scheme: A fault monitoring method of a wind turbine generator includes: s1, collecting mechanical sensor data, environmental parameter data and running state data in a wind turbine generator as an initial data set; S2, preprocessing data in an initial data set by adopting a data normalization and parameter online updating method to obtain a target data set; s3, carrying out dynamic hypergraph modeling on the operation data of the wind turbine generator according to the target data set, and capturing data relation and space-time variation; s4, constructing space-time hypergraph convolution by adopting a three-layer hypergraph convolution mode, and realizing feature aggregation; s5, performing anomaly detection by adopting a reconstruction error calculation mode and a mode of classifying and aggregating unit anomaly characteristics; s6, dynamically adjusting an alarm threshold value of the wind turbine generator based on the real-time operation condition of the wind turbine generator. The invention further improves in S1, the acquisition of mechanical sensor data, environmental parameter data and running state data in the wind turbine generator set as an initial data set comprises the steps of capturing and triggering data acquisition actions on microsecond-level time scale by utilizing high-speed parallel processing capacity based on a hardware trigger synchronous circuit of a programmable gate array, matching with a Beidou dual-mode high-precision clock timing module, internally arranging an atomic clock-level time reference, realizing nanosecond timing precision, realizing real-time caching and dynamic rearrangement by a hardware pipeline mechanism, ensuring that three source data of machinery, environment and running are aligned on a time sequence, and further taking the obtained data as the initial data set. In S2, preprocessing the data in the initial data set by adopting a data normalization and parameter online updating method to obtain a target data set, wherein the method comprises the following steps of: eliminating dimension differences among features of the initial data set by adopting a data normalization mode; normalization maps data to the same scale uniformly, and the dynamic normalization formula is as follows: Wherein, the As a result of the normalization of the characteristic values,For raw input feature values, i.e. raw sensor data from the initial dataset,Is the mean value in the K-th case,The standard deviation in the K-th updated case; The method for adjusting the mean value and standard deviation parameters in the normalization process by adopting the online parameter updating mode comprises the following steps: the self-adaptive capacity of the wind turbine generator to different working conditions is improved by adopting an online parameter updating mode, and an online parameter updating formula is as follows: Wherein, the The updated mean value under the K-th working condition,The mean value under the K-t