CN-121980237-A - Fault discrimination method and system for wind turbine generator

CN121980237ACN 121980237 ACN121980237 ACN 121980237ACN-121980237-A

Abstract

The invention relates to a fault judging method and system for a wind turbine, wherein core electrical parameters and auxiliary environmental parameters in the wind turbine are collected to serve as initial data sets, data in the initial data sets are preprocessed by adopting a data outlier judging and data aligning method to obtain target data sets, noise reduction is conducted on the target data sets by adopting an improved least mean square algorithm to obtain noise reduction data, time-frequency domain characteristics are extracted from the noise reduction data by adopting a time-frequency domain characteristic extracting algorithm to obtain aligned time-frequency characteristics, fault judgment is conducted on the time-frequency characteristics by adopting a dynamic weight generator based on a long-short term memory network, and fault judgment is conducted on the overall characteristic data of the wind turbine to obtain a predicted fault type corresponding to a power grid where the target wind turbine is located. According to the method, the influence of factors such as data acquisition, noise reduction processing and time domain feature fusion is fully considered, the reliability degree of the result is defined, and the effect of accurately reflecting the fault state of the wind turbine generator is achieved.

Inventors

Yang Zengru
CHEN BOMING
Qu Zuge
Liang Zaizhong
DU MINGHUI

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 judging method for the wind turbine generator is characterized by comprising the following steps of: s1, acquiring core electrical parameters and auxiliary environment parameters in a wind turbine generator as an initial data set; S2, preprocessing data in the initial data set by adopting a data outlier judging and data aligning method to obtain a target data set; S3, denoising the target data set by adopting an improved least mean square algorithm to obtain denoising data; s4, extracting time-frequency domain features of the noise reduction data by adopting a time-frequency domain feature extraction algorithm to obtain aligned time-frequency features; S5, performing fault characteristic enhancement on the time frequency characteristic based on a dynamic weight generator of the long-period memory network; And S6, performing fault discrimination processing based on the overall characteristic data of the wind turbine generator to obtain a predicted fault type corresponding to the power grid where the target wind turbine generator is located.
2. The method for determining a failure of a wind turbine according to claim 1, wherein in S1, collecting a core electrical parameter and an auxiliary environmental parameter in the wind turbine as an initial data set includes: The core electrical parameters comprise comprehensive and accurate acquisition of the core electrical parameters and auxiliary environmental parameters through the sensors, and abundant and reliable original data are provided for fault discrimination of the wind turbine generator, so that electrical characteristics and environmental interference information of power grid faults are more accurately captured, and the acquired data information is used as an initial data set.
3. The method for judging faults of a wind turbine generator set according to claim 2, wherein in S2, preprocessing data in an initial data set by adopting a method of judging abnormal values of data and aligning the data to obtain a target data set comprises the step of identifying abnormal values of the data in the initial data set by using a3 sigma principle based on a statistical principle; And carrying out data alignment on the initial data set after the abnormal value identification by adopting a multi-source data space-time alignment mode, and carrying out normalization processing on the integrated data matrix by adopting Min-Max normalization.
4. The method for judging faults for wind turbines according to claim 1, wherein in S3, noise reduction is carried out on a target data set by adopting an improved least mean square algorithm, noise reduction data is obtained by setting an environmental noise value, environmental noise in the target data set is eliminated, and an improved variable step-size least mean square algorithm adaptive filter is adopted, wherein a filtering formula is as follows: Wherein, the Representing the filtering error at the kth instant, i.e. the difference between the original signal and the filter output signal; Is pretreated with Is a single-parameter time-domain sequence of the sequence, The output of the filter at the kth instant.
5. The method for judging faults of a wind turbine generator set according to claim 1, wherein in the step S4, a time-frequency domain feature is extracted from noise reduction data by adopting a time-frequency domain feature extraction algorithm, and the time-frequency feature after alignment is obtained comprises the steps of extracting time domain features from a window of the noise reduction data by adopting an improved Huber regression model: The formula of the Huber loss function is as follows: Wherein, the As a Huber loss function with delta as a parameter, Delta is the threshold parameter of the Huber loss function, which is the difference between the residual, i.e., model predicted value, and the true value.
6. The fault judging method for the wind turbine generator set according to claim 1 is characterized in that in S5, a dynamic weight generator based on a long-short-term memory network LSTM is used for enhancing time-frequency characteristics, wherein the time-frequency domain characteristic information is input into a model, the LSTM layer learns dynamic change trend of the time-frequency characteristics on a time sequence through a memory unit and a gating mechanism, importance evolution rules of the time-frequency characteristics at different stages in the fault development process are mined, the full-connection layer maps the characteristics output by the LSTM layer to weights alpha, beta and gamma, the three weights are used for characteristic fusion and respectively represent contribution degrees of time-domain characteristics, frequency-domain characteristics and cross characteristics, and a weight constraint formula is as follows: Where α is a positive domain feature weight, β is a frequency domain feature weight, and γ is a cross feature weight.
7. The utility model provides a trouble discrimination system for wind turbine generator system which characterized in that includes: the data acquisition unit is used for acquiring core electrical parameters and auxiliary environment parameters in the wind turbine generator as an initial data set; the data preprocessing unit is used for preprocessing data in the initial data set by adopting a data outlier judging and data aligning method to obtain a target data set; the data denoising unit is used for denoising the target data set by adopting an improved least mean square algorithm to obtain denoising data; The time-frequency domain feature extraction unit is used for extracting time-frequency domain features from the noise reduction data by adopting a time-frequency domain feature extraction algorithm to obtain aligned time-frequency features; The fault characteristic enhancement unit is used for enhancing the fault characteristic of the time-frequency characteristic based on a dynamic weight generator of the long-period memory network; and the fault judging unit is used for carrying out fault judging processing based on the overall characteristic data of the wind turbine generator to obtain a predicted fault type corresponding to the power grid where the target wind turbine generator is located.
8. The system according to claim 7, wherein the data acquisition unit, for acquiring the core electrical parameter and the auxiliary environmental parameter in the wind turbine as the initial data set, comprises: The core electrical parameters comprise comprehensive and accurate acquisition of the core electrical parameters and auxiliary environmental parameters through the sensors, and abundant and reliable original data are provided for fault discrimination of the wind turbine generator, so that electrical characteristics and environmental interference information of power grid faults are more accurately captured, and the acquired data information is used as an initial data set.
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 discrimination method for the wind turbine generator set according to any one of claims 1 to 6.
10. A computer-readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the failure determination method for a wind turbine generator set according to any one of claims 1 to 6.

Description

Fault discrimination method and system for wind turbine generator Technical Field The invention relates to the technical field of data processing, in particular to a fault judging method and system for a wind turbine generator. Background The wind power generation industry is a renewable energy industry which utilizes natural wind energy and converts kinetic energy of wind into electric energy through a wind power generator, has the advantages of cleanness, reproducibility, low carbon, environmental protection and the like, and has important significance in optimizing an energy structure, reducing carbon emission, promoting energy transformation and sustainable development. At present, a judging method of a wind turbine generator is generally based on original data such as vibration, current or temperature collected by a sensor, collected data is subjected to threshold value setting in advance, if the data exceeds the threshold value, whether the fault occurs or not is judged, threshold value range faults are set in advance in a classifier, and fault types are judged according to the current data. However, in the actual operation process of the wind turbine generator, frequency drift can be caused by mechanical vibration or electromagnetic interference in the operation process, if the frequency of the mechanical vibration is not processed, the noise reduction effect can be affected, so that the time domain smoothness is reduced, the obtained data are not further fused by the current time-frequency domain processing, the false alarm condition is easy to occur, and the accurate fault state of the wind turbine generator is difficult to obtain. Disclosure of Invention Based on the problems in the prior art, the invention aims to provide a fault judging 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 discrimination method for a wind turbine generator comprises the following steps: s1, acquiring core electrical parameters and auxiliary environment parameters in a wind turbine generator as an initial data set; S2, preprocessing data in the initial data set by adopting a data outlier judging and data aligning method to obtain a target data set; S3, denoising the target data set by adopting an improved least mean square algorithm to obtain denoising data; s4, extracting time-frequency domain features of the noise reduction data by adopting a time-frequency domain feature extraction algorithm to obtain aligned time-frequency features; S5, performing fault characteristic enhancement on the time frequency characteristic based on a dynamic weight generator of the long-period memory network; And S6, performing fault discrimination processing based on the overall characteristic data of the wind turbine generator to obtain a predicted fault type corresponding to the power grid where the target wind turbine generator is located. The invention further improves that in S1, collecting the core electrical parameters and the auxiliary environment parameters in the wind turbine generator as the initial data set comprises: The core electrical parameters comprise comprehensive and accurate acquisition of the core electrical parameters and auxiliary environmental parameters through the sensors, and abundant and reliable original data are provided for fault discrimination of the wind turbine generator, so that electrical characteristics and environmental interference information of power grid faults are more accurately captured, and the acquired data information is used as an initial data set. In S2, preprocessing data in an initial data set by adopting a data outlier judging and data aligning method, wherein obtaining a target data set comprises the step of carrying out outlier identification on the data in the initial data set by using a 3 sigma principle based on a statistical principle; And carrying out data alignment on the initial data set after the abnormal value identification by adopting a multi-source data space-time alignment mode, and carrying out normalization processing on the integrated data matrix by adopting Min-Max normalization. The invention further improves that the noise reduction is carried out on the target data set by adopting an improved least mean square algorithm, and the noise reduction data comprises the steps of eliminating the environmental noise in the target data set by setting an environmental noise value, adopting an improved variable step-size least mean square algorithm self-adaptive filter, and adopting a filtering formula as follows: Wherein, the Representing the filtering error at the kth instant, i.e. the difference between the original signal and the filter output signal; Is pretreated with Is a single-parameter time-domain sequence of the sequence,The output of the filter at the kth instant. The invention further improves that the time-frequency domain feature extraction algorithm is