Gearbox anomaly detection method and device, storage medium, equipment and program product

Abstract

The invention relates to a gearbox anomaly detection method and device, a storage medium, equipment and a program product, relates to the technical field of wind turbine generators, and improves the reliability of an anomaly detection result. The method comprises the following steps: acquiring historical operation data of the wind turbine generator in a normal operation state; inputting the historical operation data into an oil temperature prediction model to obtain an oil temperature prediction value, output by the oil temperature prediction model, of the gearbox in the wind turbine generator, the oil temperature prediction model is used for outputting an oil temperature prediction value in the following mode: extracting a first spatial-temporal feature from historical operation data from a time dimension and a space dimension at the same time through a feature extraction layer, and extracting a second spatial-temporal feature from the historical operation data from the time dimension and the space dimension in sequence, obtaining a space-time fusion feature according to the first space-time feature and the second space-time feature, and performing oil temperature prediction through a prediction layer according to the space-time fusion feature; and determining whether the gearbox is in an abnormal state or not according to the oil temperature predicted value.

Description

Technical Field

[0001] The present disclosure relates to the technical field of wind turbines, and in particular to a gearbox abnormality detection method, device, storage medium, equipment and program product. Background Art

[0002] Wind turbines are usually located in remote mountains, near or far seas, and in "sand desert" areas, and the nacelles are usually installed tens or even hundreds of meters above the ground (sea level). The operating environment is extremely harsh, and the operating conditions are complex and changeable, making the daily operation and maintenance of wind turbines extremely difficult. The gearbox, as a key device in the wind turbine transmission chain system, plays an extremely important role in the process of the wind turbine converting wind energy into rotational mechanical energy, and the abnormal state of the gearbox needs to be detected.

[0003] At present, the oil temperature prediction model constructed by the recurrent neural network is used to mine the time series characteristics of the wind turbine SCADA (Supervisory Control And Data Acquisition) data, determine the gearbox oil temperature prediction value, and perform anomaly detection based on the gearbox oil temperature prediction value. However, because the oil temperature prediction value only considers the time series characteristics of the SCADA data, it cannot take into account the spatial characteristics of the SCADA data, resulting in unreliable oil temperature prediction values, which in turn affects the gearbox anomaly detection results. Summary of the invention

[0004] In order to solve the deficiencies of the prior art, the present disclosure provides a gearbox abnormality detection method, device, storage medium, equipment and program product.

[0005] In order to achieve the above objectives, in a first aspect, the present disclosure provides a gearbox abnormality detection method, comprising: Obtain historical operating data of wind turbines under normal operating conditions; Input the historical operation data into the oil temperature prediction model to obtain the oil temperature prediction value of the gearbox in the wind turbine set output by the oil temperature prediction model, wherein the oil temperature prediction model is used to output the oil temperature prediction value in the following manner: extracting a first spatiotemporal feature from the historical operation data from both the time dimension and the space dimension through a feature extraction layer, and extracting a second spatiotemporal feature from the historical operation data from both the time dimension and the space dimension in sequence, obtaining a spatiotemporal fusion feature based on the first spatiotemporal feature and the second spatiotemporal feature, and performing oil temperature prediction based on the spatiotemporal fusion feature through a prediction layer; According to the predicted oil temperature value, it is determined whether the gearbox is in an abnormal state.

[0006] In a second aspect, the present disclosure provides a gearbox abnormality detection device, comprising: An acquisition module is configured to acquire historical operation data of the wind turbine generator set under normal operation; A prediction module is configured to input the historical operation data into an oil temperature prediction model to obtain an oil temperature prediction value of the gearbox in the wind turbine set output by the oil temperature prediction model, wherein the oil temperature prediction model is used to output the oil temperature prediction value in the following manner: extracting a first spatiotemporal feature from the historical operation data from both a time dimension and a space dimension through a feature extraction layer, and extracting a second spatiotemporal feature from the historical operation data from both a time dimension and a space dimension in sequence, obtaining a spatiotemporal fusion feature based on the first spatiotemporal feature and the second spatiotemporal feature, and performing oil temperature prediction based on the spatiotemporal fusion feature through a prediction layer; The execution module is configured to determine whether the gearbox is in an abnormal state according to the oil temperature prediction value.

[0007] In a third aspect, the present disclosure provides a non-temporary computer-readable storage medium having a computer program stored thereon, which implements the method described in the first aspect when executed by a processor.

[0008] In a fourth aspect, the present disclosure provides an electronic device, including: a memory having a computer program stored thereon; A processor is used to execute the computer program in the memory to implement the method described in the first aspect.

[0009] In a fifth aspect, the present disclosure provides a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0010] Through the above technical solution, the oil temperature prediction model outputs the oil temperature prediction value according to the historical operation data, and determines whether the gearbox is in an abnormal state according to the oil temperature prediction value. Because the feature extraction layer in the oil temperature prediction model extracts the spatiotemporal fusion features from the historical operation data from the time dimension and the space dimension, the prediction layer of the oil temperature prediction model predicts the oil temperature according to the spatiotemporal fusion features, and the obtained oil temperature prediction value takes into account the spatiotemporal characteristics of the historical operation data, thereby ensuring the accuracy and reliability of the prediction results, and further improving the reliability of the abnormal identification results.

[0011] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings: Figure 1The figure is a flow chart of a gearbox abnormality detection method according to an exemplary embodiment of the present disclosure.

[0013] Figure 2 is another flow chart of a gearbox abnormality detection method according to an exemplary embodiment of the present disclosure.

[0014] Figure 3 is another flow chart of a gearbox abnormality detection method according to an exemplary embodiment of the present disclosure.

[0015] Figure 4 The invention is a block diagram of a gearbox abnormality detection device according to an exemplary embodiment of the present invention.

[0016] Figure 5 is a block diagram of an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0017] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0018] It should be noted that all actions of acquiring signals, information or data in the present disclosure are carried out in compliance with the relevant data protection laws and policies of the country where the device is located and with the authorization given by the owner of the corresponding device.

[0019] It should be understood that wind energy is one of the clean and renewable energy sources with the greatest prospects for large-scale application, wind power generation is one of the most mature green power generation technologies, and vigorously developing wind energy (wind power generation) is an effective way to deal with the three major problems of energy shortage, environmental pollution, and climate change. Wind turbines are the main equipment for achieving wind power generation.

[0020] The internal structure of the gearbox in a wind turbine is complex, including multiple components such as shafts, bearings and gears. In addition, the wind turbine operates for a long time under complex working conditions such as impact loads caused by gusts or turbulence or irregular dynamic alternating loads, resulting in a high failure rate of the wind turbine and a long downtime for a single failure, which seriously affects the power generation and economic benefits of the wind farm.

[0021] At present, the fault diagnosis of wind turbines can be carried out based on the analysis results of the gearbox vibration signal in the time domain, frequency domain and time-frequency domain, combined with expert experience. It is also possible to build a fault diagnosis model based on the characteristics of the gearbox vibration signal in the time-frequency domain or the modal decomposition results, through traditional machine learning algorithms or shallow neural networks, and use this fault diagnosis model to diagnose wind turbines. It is also possible to build a normal behavior model of gearbox oil temperature (such as a gearbox oil temperature prediction model) based on the SCADA time series data of the wind turbine, using a recurrent neural network (such as Recurrent Neural Network, RNN; Long Short-Term Memory, LSTM; Gated Recurrent Unit, GRU, etc.), and realize automatic identification of the early degradation state of the gearbox based on the residual change trend between the predicted value and the measured value of the gearbox oil temperature.

[0022] However, the inventors found that the expert experience fault diagnosis method based on the time-frequency domain characteristics of the gearbox vibration signal requires certain specialized knowledge and is limited by expert experience, resulting in certain subjectivity and limitations in the diagnosis results. Traditional machine learning algorithms or shallow neural networks have weak learning capabilities and cannot effectively mine the fault characteristics contained in the gearbox vibration signal. In addition, the acquisition of gearbox vibration signals requires the installation of additional hardware equipment such as collection, transmission and storage, which increases the installation and operation and maintenance costs of wind farms. The massive high-dimensional, multi-source heterogeneous, and dynamically time-varying SCADA time series data of wind turbines essentially has dual attributes of time and space. Recurrent neural networks can only mine SCADA time series features, and have certain limitations in learning SCADA spatial features, resulting in the diagnosis process being unable to take into account the spatial characteristics of SCADA data and the reliability of the diagnosis results being unable to be guaranteed.

[0023] In view of this, the present disclosure provides a gearbox anomaly detection method, device, storage medium, equipment and program product, which can take into account the temporal and spatial characteristics of the data, thereby ensuring the accuracy and reliability of the prediction results, and further improving the reliability of the anomaly identification results.

[0024] Figure 1 1 is a flow chart of a gearbox abnormality detection method according to an exemplary embodiment of the present disclosure. The method can be used in smart terminal devices such as laptops, smart phones, etc. Figure 1 As shown, the method may include the following steps: In step S11, historical operation data of the wind turbine generator system in normal operation is obtained.

[0025] It is worth noting that in order to facilitate the wind farm to monitor, control, operate and maintain the operating status of the wind turbine, a general wind farm is deployed with a data acquisition and monitoring control system (SCADA). The SCADA system can collect, store and statistically analyze the operating status parameters (such as temperature, pressure, speed, torque, power, current, voltage, etc.) and environmental parameters (such as wind speed, wind direction, temperature, air pressure, humidity, etc.) of the key equipment of the wind turbine in real time based on different sampling frequencies (such as seconds, minutes, etc.). SCADA time series data has the advantages of massive, high-dimensional, multi-source, and heterogeneous, and has excellent economy and availability. Therefore, the historical operation data in the embodiment of the present disclosure is preferably SCADA historical data. Based on SCADA time series data, the state of the wind turbine gearbox is monitored by artificial intelligence algorithms such as machine learning and deep learning to realize the automatic identification of the early degradation state of the gearbox, which is to improve the reliability and safety of wind turbine operation, reduce the unplanned downtime of the unit, and reduce the operation and maintenance cost of the unit.

[0026] It should be understood that the SCADA historical data collected under normal operating conditions of wind turbines can be cleaned using the Thompson-tau method (Thompson sampling method) to remove abnormal data in the SCADA historical data caused by wind power curtailment, operating condition fluctuations, data collection equipment or communication line failures, etc., to obtain SCADA historical health data, and then the cleaned SCADA historical health data can be normalized to eliminate the numerical differences in different state parameters in the SCADA historical health data.

[0027] In step S12, the historical operation data is input into the oil temperature prediction model to obtain the oil temperature prediction value of the gearbox in the wind turbine output by the oil temperature prediction model, wherein the oil temperature prediction model is used to output the oil temperature prediction value in the following manner: extracting a first spatiotemporal feature from the historical operation data from both the time dimension and the space dimension through a feature extraction layer, and extracting a second spatiotemporal feature from the historical operation data from both the time dimension and the space dimension in turn, obtaining a spatiotemporal fusion feature based on the first spatiotemporal feature and the second spatiotemporal feature, and performing oil temperature prediction based on the spatiotemporal fusion feature through a prediction layer.

[0028] For example, the SCADA historical health data is input into the oil temperature prediction model, and the oil temperature prediction model extracts the first spatiotemporal feature from the SCADA historical health data from both the time dimension and the space dimension through a feature extraction layer, and extracts the second spatiotemporal feature from the SCADA historical health data from both the time dimension and the space dimension in turn, and obtains a spatiotemporal fusion feature based on the first spatiotemporal feature and the second spatiotemporal feature. The oil temperature is predicted based on the spatiotemporal fusion feature through the prediction layer, and the oil temperature prediction value of the gearbox in the wind turbine set output by the oil temperature prediction model is obtained.

[0029] In step S13, it is determined whether the gearbox is in an abnormal state according to the predicted oil temperature value.

[0030] In the disclosed embodiment, the oil temperature prediction model outputs the oil temperature prediction value according to the historical operation data, and determines whether the gearbox is in an abnormal state according to the oil temperature prediction value. Because the feature extraction layer in the oil temperature prediction model extracts the spatiotemporal fusion features from the historical operation data in the time dimension and the space dimension, the prediction layer of the oil temperature prediction model predicts the oil temperature according to the spatiotemporal fusion features, and the obtained oil temperature prediction value takes into account the spatiotemporal features of the historical operation data, thereby ensuring the accuracy and reliability of the prediction results, and further improving the reliability of the abnormal identification results.

[0031] In order to facilitate those skilled in the art to better understand the gearbox abnormality detection method provided by the present disclosure, the steps of the method are described in detail below.

[0032] In a feasible implementation manner, before inputting the historical operation data into the oil temperature prediction model, the method further includes: performing a correlation analysis on the historical operating data according to a preset correlation coefficient, and selecting first target operating data from the historical operating data, wherein the correlation between the first target operating data and the gearbox oil temperature is higher than the correlation between the remaining data in the historical operating data except the first target operating data and the gearbox oil temperature; and / or, Respectively performing time attention weighted processing and space attention weighted processing on the historical operation data to enhance the time variation characteristics and space variation characteristics of the historical operation data to obtain second target operation data; The step of inputting the historical operation data into the oil temperature prediction model comprises: The first target operating data and / or the second target operating data are input into an oil temperature prediction model.

[0033] It is worth noting that before the SCADA historical health data is input into the oil temperature prediction model, the SCADA historical health data can be processed as follows: 1. Perform correlation analysis on the SCADA historical health data according to the preset correlation coefficient, and screen out multiple SCADA data with a high degree of correlation with the gearbox oil temperature; 2. Perform time attention weighted processing and space attention weighted processing on the SCADA historical health data to enhance the time change characteristics and spatial change characteristics of the SCADA historical health data; 3. While performing correlation analysis on the SCADA historical health data according to the preset correlation coefficient, perform time attention weighted processing and space attention weighted processing on the SCADA historical health data.

[0034] For example, for the third type of processing, a correlation analysis is performed on the SCADA historical health data according to a preset correlation coefficient to determine the degree of correlation between each data in the SCADA historical health data and the gearbox oil temperature, and multiple SCADA data with a higher degree of correlation with the gearbox oil temperature are selected, and then attention weighted processing is performed on the multiple SCADA data from the time dimension and the space dimension, and attention weights are dynamically and adaptively assigned to each SCADA data, thereby enhancing the time variation characteristics and spatial variation characteristics of each SCADA data, and obtaining the target SCADA data, and inputting the target SCADA data into the oil temperature prediction model to obtain the oil temperature prediction value output by the oil temperature prediction model.

[0035] In the disclosed embodiment, by performing correlation analysis on the SCADA historical health data based on a preset correlation coefficient, multiple SCADA data with a high degree of correlation with the gearbox oil temperature are screened out, thereby reducing the amount of calculation for oil temperature prediction by the subsequent oil temperature prediction model. By performing time attention weighted processing and space attention weighted processing on the SCADA historical health data, the subsequent oil temperature prediction model can focus on key features in the SCADA historical health data when performing oil temperature prediction, and can weaken or discard redundant features in the SCADA historical health data, thereby improving the reliability and accuracy of the results of oil temperature prediction by the oil temperature prediction model.

[0036] In a feasible implementation manner, the preset correlation coefficient includes at least one of a Pearson correlation coefficient, a Spearman correlation coefficient, and a Kendall correlation coefficient; The performing correlation analysis on the historical operation data according to the preset correlation coefficient and selecting the first target operation data from the historical operation data includes: According to at least one of the Pearson correlation coefficient, the Spearman correlation coefficient and the Kendall correlation coefficient, correlation analysis is performed on the historical operation data, and first target operation data is screened out from the historical operation data.

[0037] It is worth noting that in the embodiment of the present disclosure, it is preferred to perform a mixed correlation analysis on the SCADA historical health data based on the Pearson correlation coefficient, the Spearman correlation coefficient and the Kendall correlation coefficient at the same time, thereby improving the reliability of the correlation degree between each data in the determined SCADA historical health data and the gearbox oil temperature, so as to obtain SCADA data with a high degree of correlation with the wind turbine gearbox oil temperature as the input variable of the subsequent model.

[0038] In a feasible implementation manner, performing time attention weighted processing and space attention weighted processing on the historical operation data respectively, enhancing the time variation characteristics and space variation characteristics of the historical operation data, and obtaining the second target operation data, includes: Sliding a preset sliding window according to the historical operation data until the preset sliding window covers the historical operation data; Before sliding the preset sliding window for the first time and after sliding the preset sliding window each time, time attention weighted processing and space attention weighted processing are performed on the data in the preset sliding window to obtain second target operation data.

[0039] It is worth noting that the sliding step size of the preset sliding window can be preset according to the attention weighted processing speed or accuracy, and the present disclosure does not limit this.

[0040] In the disclosed embodiment, a preset sliding window is set according to the SCADA data sliding, and time attention weighted processing and space attention weighted processing are performed on the data in the preset sliding window. This can reduce the interference of the time characteristics of each data on the attention weighted processing on the basis of improving the efficiency of the time attention weighted processing and the space attention weighted processing, that is, reduce the time complexity, thereby improving the reliability of the results of the time attention weighted processing and the space attention weighted processing.

[0041] For example, Figure 2 As shown, the gearbox abnormality detection method may further include the following steps: In step S21, the SCADA historical data of the wind turbine generator system under normal operation is obtained.

[0042] In step S22, a correlation analysis is performed on the SCADA historical data according to a preset correlation coefficient, and multiple SCADA data are screened out from the SCADA historical data, wherein the correlation degree between the multiple SCADA data and the gearbox oil temperature is higher than the correlation degree between the remaining data in the historical operation data except the SCADA data and the gearbox oil temperature.

[0043] In step S23, time attention weighted processing and space attention weighted processing are performed on the SCADA data respectively to enhance the time variation characteristics and space variation characteristics of the SCADA data to obtain target SCADA data.

[0044] In step S24, the target SCADA data is input into the oil temperature prediction model to obtain the predicted value of the oil temperature of the gearbox in the wind turbine set output by the oil temperature prediction model.

[0045] In step S25, it is determined whether the gearbox is in an abnormal state according to the predicted oil temperature value.

[0046] In the disclosed embodiment, by performing correlation analysis, time attention weighted processing and space attention weighted processing on the acquired SCADA historical data in sequence, target SCADA data is obtained, and the target SCADA data is input into the oil temperature prediction model to obtain the oil temperature prediction value, which can reduce the amount of calculation for oil temperature prediction by the subsequent oil temperature prediction model and improve the reliability and accuracy of the results of oil temperature prediction by the oil temperature prediction model.

[0047] In a feasible implementation manner, the feature extraction layer of the oil temperature prediction model includes a spatial feature extraction module and a temporal feature extraction module, the first spatiotemporal feature is extracted from the historical operation data from the temporal dimension and the spatial dimension at the same time, and the second spatiotemporal feature is extracted from the historical operation data from the temporal dimension and the spatial dimension in sequence, and the spatiotemporal fusion feature is obtained according to the first spatiotemporal feature and the second spatiotemporal feature, including: Extracting a first spatial feature of the historical operation data by the spatial feature extraction module, extracting a first temporal feature of the historical operation data by the temporal feature extraction module, and fusing the first spatial feature and the first temporal feature into a first spatiotemporal feature; For the historical operation data after the first spatial feature is extracted, extracting a second time feature through the time feature extraction module, and fusing the first spatial feature and the second time feature into a second spatiotemporal feature; The first spatiotemporal feature and the second spatiotemporal feature are concatenated to obtain a spatiotemporal fusion feature.

[0048] For example, the oil temperature prediction model extracts the first spatial feature of the SCADA historical health data through the spatial feature extraction module through the feature extraction layer, extracts the first time feature of the SCADA historical health data through the time feature extraction module, and fuses the first spatial feature and the first time feature into a parallel space-time feature; for the SCADA historical health data after extracting the first spatial feature, the second time feature is extracted through the time feature extraction module, the first spatial feature and the second time feature are fused into a serial space-time feature, and the parallel space-time feature and the serial space-time feature are spliced ​​to obtain the space-time fusion feature.

[0049] In the disclosed embodiment, the parallel spatiotemporal features of spatial feature mining and temporal feature mining of SCADA historical health data are simultaneously integrated with the serial spatiotemporal features of first mining spatial features of SCADA historical health data and then mining temporal features to obtain spatiotemporal fusion features. The spatiotemporal fusion features obtained by acquiring and fusing two types of spatiotemporal features from different angles can deeply mine the spatiotemporal features contained in the SCADA historical health data, so that the oil temperature prediction value obtained according to the spatiotemporal fusion features can take into account the spatiotemporal characteristics of the SCADA historical health data at the same time, thereby improving the accuracy and reliability of the oil temperature prediction value.

[0050] It is worth noting that the prediction layer can be a fully connected neural network (FC), which is not limited in the present disclosure.

[0051] In a feasible implementation, the spatial feature extraction module includes a multi-scale convolutional neural network, and the temporal feature extraction module includes a bidirectional simple cycle unit network.

[0052] It is worth noting that the Multi-Scale Convolutional Neural Network (MSCNN) has the characteristics of strong feature representation ability, high robustness, and good generalization performance. In this embodiment, the Multi-Scale Convolutional Neural Network uses convolution kernels of different sizes to realize the mining of multi-level spatial features, ensuring the accuracy and reliability of the obtained spatial features. The Bidirectional Simple Recurrent Unit (BiSRU) has the characteristics of high computational efficiency, strong parallel processing capability, and easy implementation. It can be obtained by improving the Bidirectional Recurrent Neural Network (RNN). In the embodiment of the present disclosure, the Bidirectional Simple Recurrent Unit (BiSRU) can effectively capture the bidirectional dependency relationship in the sequence, and has high computational efficiency and the ability to adapt to long sequence processing, which improves the accuracy and reliability of the obtained time features.

[0053] In the embodiments of the present invention, a multi-scale convolutional neural network and a bidirectional recurrent neural network are combined to extract spatiotemporal features of SCADA data, which can deeply mine the spatiotemporal characteristics contained in the SCADA data, obtain spatiotemporal fusion features that can effectively characterize the law of gearbox oil temperature changes, and thus improve the accuracy and reliability of gearbox oil temperature prediction.

[0054] In a feasible implementation manner, determining whether the gearbox is in an abnormal state according to the predicted oil temperature value includes: Determining the oil temperature value of the gearbox in the wind turbine generator set; Determining the oil temperature prediction residual of the gearbox according to the oil temperature value and the oil temperature prediction value; When the oil temperature prediction residual is greater than a preset residual threshold, determining that the gearbox is in an abnormal state; When the oil temperature prediction residual is less than or equal to the preset residual threshold, it is determined that the gearbox is in a normal state.

[0055] It should be understood that determining the oil temperature value of the gearbox in the wind turbine can be achieved by online detection, which can specifically include the following steps: obtaining the SCADA online data of the wind turbine, inputting the SCADA online data into a trained oil temperature prediction model, and obtaining the oil temperature value of the gearbox in the wind turbine. In the disclosed embodiment, the oil temperature of the gearbox in the wind turbine can be predicted by a trained oil temperature prediction model based on the SCADA online data of the wind turbine.

[0056] It is worth noting that the oil temperature prediction model in the embodiment of the present disclosure can be obtained by offline training with the above-mentioned target SCADA data, the above-mentioned target SCADA data is divided into a training set and a test set, the training is input into the oil temperature prediction model, and the oil temperature prediction value of the gearbox in the wind turbine set output by the oil temperature prediction model is obtained. And the preset residual threshold is determined according to the test set.

[0057] Among them, the preset residual threshold can be determined in the following way: determine the oil temperature prediction residual corresponding to each data in the test set, statistically analyze the distribution law of the oil temperature prediction residual corresponding to all the data in the test set by the kernel density estimation method, determine the probability density distribution of the oil temperature prediction residual under the normal operation of the wind turbine, and set the residual threshold corresponding to the gearbox oil temperature abnormality under a certain confidence level according to the probability density distribution of the oil temperature prediction residual, so as to avoid misjudgment of abnormal state caused by oil temperature jump of the gearbox.

[0058] It is worth noting that when the gearbox is in an abnormal state, the abnormal proportion of the oil temperature residual value in the sliding window can be further counted, and the cumulative sum of the absolute value of the oil temperature residual in the abnormal time period in the sliding window can be calculated. When the abnormal proportion is greater than the preset value, the gearbox oil temperature abnormal alarm is triggered, and the residual cumulative sum in the abnormal time period is used to measure the severity of the gearbox oil temperature abnormality. The preset value can be preset according to the warning accuracy of the wind turbine gearbox, and this disclosure does not limit this.

[0059] For example, Figure 3 As shown, the gearbox anomaly detection method includes two stages: offline training and online monitoring.

[0060] Offline training includes the following steps: I. Data collection and data preprocessing: Collect the SCADA historical data of wind turbines under normal operating conditions, and clean them using the Thompson-tau method to remove abnormal data caused by wind power curtailment, operating condition fluctuations, data collection equipment or communication line failures, and obtain SCADA historical health data; normalize the cleaned SCADA historical health data to eliminate the adverse effects of large differences in the values ​​of parameters in different states in the SCADA data.

[0061] II. Selection of model input variables: Based on the Pearson correlation coefficient, Spearman correlation coefficient and Kendall correlation coefficient, a mixed correlation analysis is performed on the SCADA historical health data to obtain the data with a high degree of correlation with the oil temperature of the wind turbine gearbox in the SCADA historical health data as the input variables of the subsequent model, thereby reducing the calculation amount of the model and improving the efficiency of model training.

[0062] III. Confirmation of model input matrix: A sliding window is used to slide the SCADA historical health data for hybrid correlation analysis to determine multiple model input matrices.

[0063] Split multiple model input matrices into training and test sets.

[0064] IV. Spatiotemporal attention weighting of model input: Each model input matrix in the training set is weighted for attention in the temporal dimension and the spatial dimension respectively, and attention weights are dynamically and adaptively assigned to the model input matrix.

[0065] V. Parallel spatiotemporal feature mining and fusion: For the model input matrix after spatiotemporal attention weighting obtained in IV, a spatial feature mining network based on a multi-scale convolutional neural network and a temporal feature mining network based on a bidirectional simple recurrent unit network are constructed respectively; the spatial feature mining network and the temporal feature mining network mine spatial features and temporal features in parallel at the same time, and the spatial features and temporal features mined are spliced ​​to obtain parallel spatiotemporal fusion features.

[0066] VI. Serial spatiotemporal feature mining and fusion: For the model input matrix after spatiotemporal attention weighting obtained in IV, a spatial feature mining network based on a multi-scale convolutional neural network and a temporal feature mining network based on a bidirectional simple recurrent unit network are sequentially constructed. The spatial features are first mined, and then the temporal features are mined to obtain the serial spatiotemporal fusion features.

[0067] VII. Serial and parallel spatiotemporal feature fusion: Based on V, parallel spatiotemporal fusion features are obtained, and based on VI, serial spatiotemporal fusion features are obtained, and then spliced ​​together to obtain spatiotemporal fusion features.

[0068] Gearbox oil temperature prediction: The spatiotemporal fusion features are used as the input of the fully connected neural network to obtain the gearbox oil temperature prediction value output by the fully connected neural network.

[0069] Probability density distribution of prediction residuals of the test set: The distribution law of the prediction residuals corresponding to the test set is statistically analyzed by the kernel density estimation method, and the probability density distribution law of the model prediction residuals under the normal operation of the wind turbine is explored and statistically analyzed to determine the probability density distribution of the oil temperature prediction residuals under the normal operation of the wind turbine. Based on the probability density distribution of the test set model prediction residuals, the abnormal threshold of the gearbox oil temperature abnormality under a certain confidence level is set, that is, the preset residual threshold.

[0070] Among them, online monitoring includes the following steps: VIII. Determine the oil temperature prediction residual: Based on the wind turbine SCADA online data, the oil temperature prediction model is used to obtain the measured value of the gearbox oil temperature according to the SCADA online data, and the predicted value is compared with the measured value to obtain the gearbox oil temperature prediction residual.

[0071] VIII. Based on the abnormal threshold obtained in VII and the oil temperature prediction residual obtained in VIII, the gearbox operation status is judged to be abnormal. When the oil temperature prediction residual exceeds the abnormal threshold, the gearbox is determined to be in an abnormal state. The abnormal state ratio corresponding to each data in the sliding window is counted, and the cumulative sum of the absolute values ​​of the prediction residuals in the abnormal time period in the sliding window is calculated to measure the severity of the gearbox oil temperature abnormality; when the abnormal ratio is greater than the preset value, the gearbox is determined to be faulty, and the gearbox oil temperature abnormality alarm is issued; when the abnormal ratio is less than or equal to the preset value, the gearbox is determined to be normal. When the oil temperature prediction residual exceeds the abnormal threshold, the gearbox is determined to be in a normal state.

[0072] The above scheme mainly includes two stages: offline training (I-VII) and online monitoring (VIII-VIIII). The offline training stage is mainly based on the historical health data of wind turbine SCADA after data cleaning and data preprocessing to complete the training and testing of the oil temperature prediction model. The online monitoring stage is mainly based on the online data of wind turbine SCADA. According to the oil temperature prediction model, the prediction residual between the predicted value and the measured value of the oil temperature is obtained, so as to monitor the change trend of the prediction residual in real time and realize the automatic identification of the gearbox degradation state.

[0073] Based on the same inventive concept, the present disclosure also provides a block diagram of a gearbox abnormality detection device, such as Figure 4 As shown, the gearbox abnormality detection device includes an acquisition module 401 , a prediction module 402 and an execution module 403 .

[0074] Wherein, the acquisition module 401 is configured to acquire historical operation data of the wind turbine generator set under normal operation state; The prediction module 402 is configured to input the historical operation data into the oil temperature prediction model to obtain the oil temperature prediction value of the gearbox in the wind turbine output by the oil temperature prediction model, wherein the oil temperature prediction model is used to output the oil temperature prediction value in the following manner: extracting a first spatiotemporal feature from the historical operation data from both the time dimension and the space dimension through a feature extraction layer, and extracting a second spatiotemporal feature from the historical operation data from both the time dimension and the space dimension in sequence, obtaining a spatiotemporal fusion feature based on the first spatiotemporal feature and the second spatiotemporal feature, and performing oil temperature prediction based on the spatiotemporal fusion feature through a prediction layer; The execution module 403 is configured to determine whether the gearbox is in an abnormal state according to the oil temperature prediction value.

[0075] In the disclosed embodiment, the oil temperature prediction model outputs the oil temperature prediction value according to the historical operation data, and determines whether the gearbox is in an abnormal state according to the oil temperature prediction value. Because the feature extraction layer in the oil temperature prediction model extracts the spatiotemporal fusion features from the historical operation data in the time dimension and the space dimension, the prediction layer of the oil temperature prediction model predicts the oil temperature according to the spatiotemporal fusion features, and the obtained oil temperature prediction value takes into account the spatiotemporal features of the historical operation data, thereby ensuring the accuracy and reliability of the prediction results, and further improving the reliability of the abnormal identification results.

[0076] Optionally, the feature extraction layer of the oil temperature prediction model includes a spatial feature extraction module and a temporal feature extraction module, and the prediction module 402 is configured to extract a first spatial feature of the historical operation data through the spatial feature extraction module, extract a first temporal feature of the historical operation data through the temporal feature extraction module, and fuse the first spatial feature and the first temporal feature into a first spatiotemporal feature; For the historical operation data after the first spatial feature is extracted, extracting a second time feature through the time feature extraction module, and fusing the first spatial feature and the second time feature into a second spatiotemporal feature; The first spatiotemporal feature and the second spatiotemporal feature are concatenated to obtain a spatiotemporal fusion feature.

[0077] Optionally, the spatial feature extraction module includes a multi-scale convolutional neural network, and the temporal feature extraction module includes a bidirectional simple cycle unit network.

[0078] Optionally, the gearbox abnormality detection device further includes a data preprocessing module, the data preprocessing module being configured to perform a correlation analysis on the historical operating data according to a preset correlation coefficient, and filter out first target operating data from the historical operating data, wherein the correlation between the first target operating data and the gearbox oil temperature is higher than the correlation between the remaining data in the historical operating data except the first target operating data and the gearbox oil temperature; and / or, Respectively performing time attention weighted processing and space attention weighted processing on the historical operation data to enhance the time variation characteristics and space variation characteristics of the historical operation data to obtain second target operation data; The prediction module is configured to input the first target operating data and / or the second target operating data into an oil temperature prediction model.

[0079] Optionally, the preset correlation coefficient includes at least one of a Pearson correlation coefficient, a Spearman correlation coefficient and a Kendall correlation coefficient; The data preprocessing module is configured to perform correlation analysis on the historical operation data according to at least one of the Pearson correlation coefficient, the Spearman correlation coefficient and the Kendall correlation coefficient, and filter out the first target operation data from the historical operation data.

[0080] Optionally, the data preprocessing module is configured to slide a preset sliding window according to the historical operation data until the preset sliding window covers the historical operation data; Before sliding the preset sliding window for the first time and after sliding the preset sliding window each time, time attention weighted processing and space attention weighted processing are performed on the data in the preset sliding window to obtain second target operation data.

[0081] Optionally, the execution module is configured to determine an oil temperature value of a gearbox in the wind turbine generator set; Determining the oil temperature prediction residual of the gearbox according to the oil temperature value and the oil temperature prediction value; When the oil temperature prediction residual is greater than a preset residual threshold, determining that the gearbox is in an abnormal state; When the oil temperature prediction residual is less than or equal to the preset residual threshold, it is determined that the gearbox is in a normal state.

[0082] Regarding the gearbox abnormality detection device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0083] Based on the same inventive concept, the present disclosure also provides an electronic device, including: a memory having a computer program stored thereon; A processor is used to execute the computer program in the memory to implement the above-mentioned gearbox abnormality detection method.

[0084] In the disclosed embodiment, the oil temperature prediction model outputs the oil temperature prediction value according to the historical operation data, and determines whether the gearbox is in an abnormal state according to the oil temperature prediction value. Because the feature extraction layer in the oil temperature prediction model extracts the spatiotemporal fusion features from the historical operation data in the time dimension and the space dimension, the prediction layer of the oil temperature prediction model predicts the oil temperature according to the spatiotemporal fusion features, and the obtained oil temperature prediction value takes into account the spatiotemporal features of the historical operation data, thereby ensuring the accuracy and reliability of the prediction results, and further improving the reliability of the abnormal identification results.

[0085] Figure 5 FIG. 5 is a block diagram of an electronic device 500 according to an exemplary embodiment. Figure 5 As shown, the electronic device 500 may include: a processor 501 , a memory 502 . The electronic device 500 may also include one or more of a multimedia component 503 , an input / output (I / O) interface 504 , and a communication component 505 .

[0086] The processor 501 is used to control the overall operation of the electronic device 500 to complete all or part of the steps in the above-mentioned gearbox abnormality detection method. The memory 502 is used to store various types of data to support the operation of the electronic device 500, and these data may include instructions for any application or method used to operate on the electronic device 500, and application-related data, such as SCADA historical data, SCADA online data, etc. The memory 502 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk. The multimedia component 503 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 502 or sent through the communication component 505. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 504 provides an interface between the processor 501 and other interface modules, and the other interface modules may be keyboards, mice, buttons, etc. These buttons may be virtual buttons or physical buttons. The communication component 505 is used for wired or wireless communication between the electronic device 500 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IOT, eMTC, or other 5G, etc., or a combination of one or more of them, is not limited here. Therefore, the corresponding communication component 505 may include: Wi-Fi module, Bluetooth module, NFC module, etc.

[0087] In an exemplary embodiment, the electronic device 500 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned gearbox abnormality detection method.

[0088] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned gearbox abnormality detection method are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 502 including program instructions, and the above-mentioned program instructions can be executed by the processor 501 of the electronic device 500 to complete the above-mentioned gearbox abnormality detection method.

[0089] In another exemplary embodiment, a computer program product is also provided. The computer program product includes a computer program executable by a programmable device, and the computer program has a code portion for executing the above-mentioned gearbox abnormality detection method when executed by the programmable device.

[0090] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings; however, the present disclosure is not limited to the specific details in the above embodiments. Within the technical concept of the present disclosure, a variety of simple modifications can be made to the technical solution of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0091] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0092] In addition, various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.

Claims

1. A gearbox abnormality detection method, characterized in that: include: Obtain historical operating data of wind turbines under normal operating conditions; Input the historical operation data into the oil temperature prediction model to obtain the oil temperature prediction value of the gearbox in the wind turbine set output by the oil temperature prediction model, wherein the oil temperature prediction model is used to output the oil temperature prediction value in the following manner: extracting a first spatiotemporal feature from the historical operation data from both the time dimension and the space dimension through a feature extraction layer, and extracting a second spatiotemporal feature from the historical operation data from both the time dimension and the space dimension in sequence, obtaining a spatiotemporal fusion feature based on the first spatiotemporal feature and the second spatiotemporal feature, and performing oil temperature prediction based on the spatiotemporal fusion feature through a prediction layer; According to the predicted oil temperature value, it is determined whether the gearbox is in an abnormal state.

2. The gearbox abnormality detection method according to claim 1, characterized in that: The feature extraction layer of the oil temperature prediction model includes a spatial feature extraction module and a temporal feature extraction module, wherein the first spatiotemporal feature is extracted from the historical operation data from the temporal dimension and the spatial dimension at the same time, and the second spatiotemporal feature is extracted from the historical operation data from the temporal dimension and the spatial dimension in sequence, and the spatiotemporal fusion feature is obtained according to the first spatiotemporal feature and the second spatiotemporal feature, including: Extracting a first spatial feature of the historical operation data by the spatial feature extraction module, extracting a first temporal feature of the historical operation data by the temporal feature extraction module, and fusing the first spatial feature and the first temporal feature into a first spatiotemporal feature; For the historical operation data after the first spatial feature is extracted, extracting a second time feature through the time feature extraction module, and fusing the first spatial feature and the second time feature into a second spatiotemporal feature; The first spatiotemporal feature and the second spatiotemporal feature are concatenated to obtain a spatiotemporal fusion feature.

3. The gearbox abnormality detection method according to claim 2, characterized in that: The spatial feature extraction module includes a multi-scale convolutional neural network, and the temporal feature extraction module includes a bidirectional simple cycle unit network.

4. The gearbox abnormality detection method according to any one of claims 1 to 3, characterized in that: Before inputting the historical operation data into the oil temperature prediction model, the method further includes: performing a correlation analysis on the historical operating data according to a preset correlation coefficient, and selecting first target operating data from the historical operating data, wherein the correlation between the first target operating data and the gearbox oil temperature is higher than the correlation between the remaining data in the historical operating data except the first target operating data and the gearbox oil temperature; and / or, Respectively performing time attention weighted processing and space attention weighted processing on the historical operation data to enhance the time variation characteristics and space variation characteristics of the historical operation data to obtain second target operation data; The step of inputting the historical operation data into the oil temperature prediction model comprises: The first target operating data and / or the second target operating data are input into an oil temperature prediction model.

5. The gearbox abnormality detection method according to claim 4, characterized in that: The preset correlation coefficient includes at least one of a Pearson correlation coefficient, a Spearman correlation coefficient and a Kendall correlation coefficient; The performing correlation analysis on the historical operation data according to the preset correlation coefficient and selecting the first target operation data from the historical operation data includes: According to at least one of the Pearson correlation coefficient, the Spearman correlation coefficient and the Kendall correlation coefficient, correlation analysis is performed on the historical operation data, and first target operation data is screened out from the historical operation data.

6. The gearbox abnormality detection method according to claim 4, characterized in that: The performing time attention weighted processing and space attention weighted processing on the historical operation data respectively, enhancing the time variation characteristics and space variation characteristics of the historical operation data, and obtaining the second target operation data, includes: Sliding a preset sliding window according to the historical operation data until the preset sliding window covers the historical operation data; Before sliding the preset sliding window for the first time and after sliding the preset sliding window each time, time attention weighted processing and space attention weighted processing are performed on the data in the preset sliding window to obtain second target operation data.

7. The gearbox abnormality detection method according to claim 1, characterized in that: The step of determining whether the gearbox is in an abnormal state according to the predicted oil temperature value includes: Determining the oil temperature value of the gearbox in the wind turbine generator set; Determining the oil temperature prediction residual of the gearbox according to the oil temperature value and the oil temperature prediction value; When the oil temperature prediction residual is greater than a preset residual threshold, determining that the gearbox is in an abnormal state; When the oil temperature prediction residual is less than or equal to the preset residual threshold, it is determined that the gearbox is in a normal state.

8. A gear box abnormality detection device, characterized in that: include: An acquisition module is configured to acquire historical operation data of the wind turbine generator set under normal operation; A prediction module is configured to input the historical operation data into an oil temperature prediction model to obtain an oil temperature prediction value of the gearbox in the wind turbine set output by the oil temperature prediction model, wherein the oil temperature prediction model is used to output the oil temperature prediction value in the following manner: extracting a first spatiotemporal feature from the historical operation data from both a time dimension and a space dimension through a feature extraction layer, and extracting a second spatiotemporal feature from the historical operation data from both a time dimension and a space dimension in sequence, obtaining a spatiotemporal fusion feature based on the first spatiotemporal feature and the second spatiotemporal feature, and performing oil temperature prediction based on the spatiotemporal fusion feature through a prediction layer; The execution module is configured to determine whether the gearbox is in an abnormal state according to the oil temperature prediction value.

9. The gear box abnormality detection device according to claim 8, characterized in that: The feature extraction layer of the oil temperature prediction model includes a spatial feature extraction module and a temporal feature extraction module. The prediction module is configured to extract a first spatial feature of the historical operation data through the spatial feature extraction module, extract a first temporal feature of the historical operation data through the temporal feature extraction module, and fuse the first spatial feature and the first temporal feature into a first spatiotemporal feature; For the historical operation data after the first spatial feature is extracted, extracting a second time feature through the time feature extraction module, and fusing the first spatial feature and the second time feature into a second spatiotemporal feature; The first spatiotemporal feature and the second spatiotemporal feature are concatenated to obtain a spatiotemporal fusion feature.

10. The gear box abnormality detection device according to claim 9, characterized in that: The spatial feature extraction module includes a multi-scale convolutional neural network, and the temporal feature extraction module includes a bidirectional simple cycle unit network.

11. The gear box abnormality detection device according to any one of claims 8 to 10, characterized in that: The gearbox abnormality detection device further includes a data preprocessing module, the data preprocessing module being configured to perform a correlation analysis on the historical operation data according to a preset correlation coefficient, and filter out first target operation data from the historical operation data, wherein the correlation between the first target operation data and the gearbox oil temperature is higher than the correlation between the remaining data in the historical operation data except the first target operation data and the gearbox oil temperature; and / or, Respectively performing time attention weighted processing and space attention weighted processing on the historical operation data to enhance the time variation characteristics and space variation characteristics of the historical operation data to obtain second target operation data; The prediction module is configured to input the first target operating data and / or the second target operating data into an oil temperature prediction model.

12. The gear box abnormality detection device according to claim 11, characterized in that: The preset correlation coefficient includes at least one of a Pearson correlation coefficient, a Spearman correlation coefficient and a Kendall correlation coefficient; The data preprocessing module is configured to perform correlation analysis on the historical operation data according to at least one of the Pearson correlation coefficient, the Spearman correlation coefficient and the Kendall correlation coefficient, and filter out the first target operation data from the historical operation data.

13. The gear box abnormality detection device according to claim 11, characterized in that: The data preprocessing module is configured to slide a preset sliding window according to the historical operation data until the preset sliding window covers the historical operation data; Before sliding the preset sliding window for the first time and after sliding the preset sliding window each time, time attention weighted processing and space attention weighted processing are performed on the data in the preset sliding window to obtain second target operation data.

14. The gear box abnormality detection device according to claim 8, characterized in that: The execution module is configured to determine the oil temperature value of the gearbox in the wind turbine; Determining the oil temperature prediction residual of the gearbox according to the oil temperature value and the oil temperature prediction value; When the oil temperature prediction residual is greater than a preset residual threshold, determining that the gearbox is in an abnormal state; When the oil temperature prediction residual is less than or equal to the preset residual threshold, it is determined that the gearbox is in a normal state.

15. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

16. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the method according to any one of claims 1 to 7.

17. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images