CN-122020274-A - Fault early warning method and system for double-device integrated wind turbine generator

CN122020274ACN 122020274 ACN122020274 ACN 122020274ACN-122020274-A

Abstract

The invention discloses a fault early-warning method and a system of a double-device integrated wind turbine generator, and relates to the technical field of wind turbine generator fault early-warning, wherein the method comprises the steps of obtaining multi-mode historical operation data of the wind turbine generator, and preprocessing to construct a sample data set; and training the double-sensor sensing network by adopting the sample data set, embedding the double-sensor sensing network trained to be converged into a control center of the wind turbine generator, and carrying out fault monitoring and early warning. The method solves the technical problems that the prior art has lag in early warning of the faults of the wind turbine generator, has insufficient diagnosis precision and is difficult to accurately identify the early faults, and achieves the technical effects of accurately and early monitoring and diagnosing the faults of the wind turbine generator, and improving the timeliness of fault early warning and the accuracy of diagnosis.

Inventors

LI TAO
ZHOU HONGGUI
Yao kan
LI LONG
REN ZILONG
Hui Longxian
JIANG XIN
SHEN JIAN
CHEN FENG

Assignees

湖南大唐先一科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251204

Claims (8)

1. The fault early warning method of the double-device integrated wind turbine generator is characterized by comprising the following steps of: acquiring multi-mode historical operation data of the wind turbine generator from a master control SCADA system, and preprocessing to construct a sample data set; An automatic encoder based on a CNN, an LSTM and an attention mechanism and an online self-adaptive Kalman filter are used for constructing a double-ware sensing network; And training the double-sensor sensing network by adopting the sample data set, embedding the double-sensor sensing network trained to be converged into a control center of the wind turbine generator, and carrying out fault monitoring and early warning.
2. The fault early warning method for a double-device integrated wind turbine generator system according to claim 1, wherein the constructing of the double-device sensing network based on CNN, LSTM, attention mechanism-based automatic encoder and on-line adaptive kalman filter comprises: Extracting data local spatial features by using CNN, capturing a dependency relationship between time sequences by using LSTM, and constructing a global perceptron by taking a light CNN-BiLSTM-attention self-encoder as a core, wherein the global perceptron is used for evaluating the whole running state of the wind turbine generator and taking the output reconstruction error as a health index; Adopting a model fusing the high-frequency vibration signals and the time sequence diagram neural network to construct a special diagnostic device for deep fault diagnosis and positioning; And connecting the global perceptron with the special diagnostor, and establishing a cooperative triggering mechanism to obtain the dual-perception network.
3. The fault early warning method for a double-device integrated wind turbine generator system according to claim 2, wherein the method is characterized in that the dependency relationship between the LSTM capturing time series is captured by using CNN to extract the local spatial characteristics of data, and the method uses a lightweight CNN-BiLSTM-attention self-encoder as a core to construct a global perceptron, and comprises the following steps: constructing an encoder formed by sequentially connecting a one-dimensional convolution layer, a maximum pooling layer, a bidirectional LSTM layer and an attention mechanism layer, and extracting key space-time characteristics from an input sequence; constructing a decoder consisting of a one-dimensional convolution layer and an up-sampling layer, and reconstructing the characteristic representation output by the encoder back to the dimension of the original input sequence; And embedding an online self-adaptive Kalman filter between the encoder and the decoder, and performing state estimation and noise filtering on the characteristic sequence output by the encoder to obtain the global perceptron.
4. The fault early warning method of a double-device integrated wind turbine generator system according to claim 1, wherein a special diagnostic device is constructed by adopting a model fusing a high-frequency vibration signal and a time sequence diagram neural network, and the fault early warning method is used for performing deep fault diagnosis and positioning and comprises the following steps: acquiring a high-frequency vibration original signal of a preset subsystem from a state monitoring device independent of a master control SCADA system, aligning the high-frequency vibration original signal with multi-mode historical operation data acquired from the master control SCADA system on a time stamp, and performing splicing and fusion on a characteristic layer to obtain a fusion characteristic vector; Based on the fusion feature vector, constructing a subsystem topological graph by taking physical connection or functional association among sensors as edges on the basis of taking each sensor measuring point in a preset subsystem as a node; Based on the subsystem topological graph, the spatial dependence among nodes is captured by utilizing a graph convolution network, and the graph convolution network is combined with a gating circulation unit to jointly learn a dynamic time sequence mode of the graph structure in the fault evolution process, so that the special diagnostor is established.
5. The method for fault pre-warning of a double-device integrated wind turbine generator system according to claim 2, wherein training the double-device sensing network by using the sample data set comprises: extracting sample data under normal working conditions from the sample data set, performing unsupervised pre-training on a self-encoder in the global perceptron, and minimizing reconstruction errors between input and output; initializing the weight of the self-encoder which is pre-trained, and performing end-to-end joint supervised training with an online self-adaptive Kalman filter and the special diagnostor which introduces label data.
6. The method for fault early warning of a double-device integrated wind turbine generator system according to claim 2, wherein establishing a cooperative triggering mechanism comprises: in the non-fault operation stage of the wind turbine, counting the distribution of reconstruction errors output by the global sensor, and establishing a health state baseline; dynamically calculating an alarm threshold value of the current time window by adopting an exponential weighted moving average method based on the health state baseline; when the real-time reconstruction error continuously exceeds the alarm threshold value for a preset number of times, judging that the state of the unit is abnormal, and automatically triggering the corresponding special diagnostic device to be put into operation according to the subsystem to which the abnormal variable belongs.
7. The fault early warning method for a double-device integrated wind turbine generator set according to claim 1, wherein the method for acquiring multi-mode historical operation data of the wind turbine generator set from a master control SCADA system and preprocessing the data comprises the following steps: deleting invalid data records of the multi-mode historical operation data in the shutdown, starting and shutdown processes and the power limiting operation state of the unit, and generating first processing data; calculating importance scores of all SCADA feature variables and unit core state parameters on the first processing data by utilizing XGBoost algorithm, and selecting feature variables with importance ranking N to form an input feature set; normalizing the input feature set.
8. A fault pre-warning system for a double-device integrated wind turbine, wherein the system is configured to implement a fault pre-warning method for a double-device integrated wind turbine according to any one of claims 1 to 7, the system comprising: the sample data set construction module is used for acquiring multi-mode historical operation data of the wind turbine generator from the master control SCADA system and preprocessing the multi-mode historical operation data to construct a sample data set; The double-sensor sensing network construction module is used for constructing a double-sensor sensing network based on CNN, LSTM, an automatic encoder based on an attention mechanism and an online self-adaptive Kalman filter; And the fault monitoring and early warning module is used for training the double-sensor sensing network by adopting the sample data set, embedding the double-sensor sensing network trained to be converged into the control center of the wind turbine generator, and carrying out fault monitoring and early warning.

Description

Fault early warning method and system for double-device integrated wind turbine generator Technical Field The invention relates to the technical field of wind turbine generator fault early warning, in particular to a fault early warning method and system for a double-device integrated wind turbine generator. Background In operation and maintenance management of the wind turbine generator, timeliness and diagnosis accuracy of fault early warning are directly related to operation safety, power generation efficiency and operation and maintenance cost control of the wind turbine generator. At present, the traditional wind turbine generator system fault early warning technology has obvious limitations that on one hand, most methods depend on multi-mode operation data of a main control SCADA system, but the data preprocessing link is not always used for effectively eliminating invalid data such as shutdown, electricity limiting and the like, and accurate screening of key features is not used, so that data quality is uneven and model input reliability is affected, on the other hand, the traditional early warning model is mostly provided with a single deep learning architecture, space-time related features of the data are difficult to efficiently mine at the same time, and an adaptive optimization mechanism is not designed for a non-steady operation environment of the wind turbine generator system, so that early fault sensitivity is insufficient, and early warning lag or false alarm and missing alarm problems often occur. In the prior art, the early warning of the faults of the wind turbine generator is lagged, the diagnosis precision is insufficient, and the technical problem of early faults is difficult to accurately identify. Disclosure of Invention The application provides a fault early warning method and system of a double-device integrated wind turbine generator, which are used for solving the technical problems that in the prior art, the fault early warning of a wind turbine generator is lagged, the diagnosis precision is insufficient, and early faults are difficult to accurately identify. In view of the above problems, the application provides a fault early warning method and system for a double-device integrated wind turbine generator. The application provides a fault early warning method of a double-device integrated wind turbine generator, which comprises the following steps: the method comprises the steps of acquiring multi-mode historical operation data of a wind turbine from a master control SCADA system, preprocessing the multi-mode historical operation data to construct a sample data set, constructing a dual-sensor sensing network based on CNN, LSTM, an automatic encoder based on an attention mechanism and an online self-adaptive Kalman filter, training the dual-sensor sensing network by adopting the sample data set, embedding the dual-sensor sensing network trained to be converged into a control center of the wind turbine, and carrying out fault monitoring and early warning. In a second aspect of the present application, a fault early warning system for a wind turbine generator set integrated by two devices is provided, the system includes: The system comprises a master control SCADA system, a sample data set construction module, a dual-sensor sensing network construction module and a fault monitoring and early warning module, wherein the master control SCADA system is used for acquiring multi-mode historical operation data of a wind turbine generator and preprocessing the multi-mode historical operation data to construct a sample data set, the dual-sensor sensing network construction module is used for constructing a dual-sensor sensing network based on CNN, LSTM, an automatic encoder based on an attention mechanism and an online self-adaptive Kalman filter, and the fault monitoring and early warning module is used for training the dual-sensor sensing network by adopting the sample data set, embedding the dual-sensor sensing network trained to be converged into a control center of the wind turbine generator and carrying out fault monitoring and early warning. One or more technical schemes provided by the application have at least the following technical effects or advantages: The method comprises the steps of acquiring multi-mode historical operation data of a wind turbine from a master control SCADA system, preprocessing the multi-mode historical operation data to construct a sample data set, constructing a dual-sensor sensing network based on CNN, LSTM, an automatic encoder based on an attention mechanism and an online self-adaptive Kalman filter, training the dual-sensor sensing network by adopting the sample data set, embedding the dual-sensor sensing network trained to be converged into a control center of the wind turbine, and carrying out fault monitoring and early warning. The method achieves the technical effects of realizing accurate early monitoring and diagnosis of faults of the wind t