Wind turbine operating status evaluation method and system based on forward-forward propagation

Through the forward-forward propagation method and ResNet network model, combined with clustering and space expansion labeling technology, the problems of low efficiency and poor accuracy of state evaluation of traditional wind turbines are solved, and more accurate and timely fault detection is achieved, which improves wind power The unit's operating status evaluation capability.

CN118532303BActive Publication Date: 2024-11-01HUNAN UNIV OF SCI & TECH
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411009105.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-01
Estimated Expiration
2044-07-26

AI Technical Summary

Technical Problem

Traditional wind turbine status evaluation methods are inefficient and have poor accuracy, making it difficult to perform predictive maintenance, and the evaluation process is limited by physical inspection and subjective judgment, making it difficult to make full use of operating data.

Method used

Using a forward-forward propagation method, the ResNet network model is used to combine clustering and spatial extension labeling methods to train the model through historical data, generate label images and perform grid search optimization, for real-time state evaluation of wind turbines.

Benefits of technology

It improves the accuracy and timeliness of wind turbine status evaluation, can automatically process large amounts of data, reduce manual intervention, enhances the accuracy and consistency of fault detection, and supports smarter maintenance decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118532303B_ABST
    Figure CN118532303B_ABST
Patent Text Reader

Abstract

The embodiments of the present disclosure provide a method and system for evaluating the operating status of a wind turbine based on forward-forward propagation. Applied to the technical fields of data processing and data prediction, it includes drawing different operating characteristic curves according to the operating data of the wind turbine, processing the different operating characteristic curves to obtain the original data set, and dividing it into a training set and a test set; generating a label image based on a spatial extension label method, and superimposing it with the training set to obtain positive and negative sample images; inputting the positive and negative sample images into a ResNet network model based on a forward-forward propagation method for training to obtain a trained network model; acquiring the operating data of the wind turbine in real time, and inputting it into the network model for predictive evaluation to obtain an evaluation result. In this way, the present disclosure utilizes the forward-forward propagation method in combination with the residual neural network to effectively capture and learn the features and label information in the data, thereby improving the timeliness and accuracy of fault detection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the technical field of data processing and data prediction, and in particular to a method and system for evaluating the operating status of a wind turbine generator system based on forward-forward propagation. Background Art

[0002] As the global demand for sustainable energy increases, wind power generation has become one of the most important renewable energy sources. It plays an increasingly important role in the global energy structure. The stable and efficient operation of wind turbines is directly related to the reliable supply of energy and the stability of the power system. The normal operation of wind turbines can not only ensure the maximum power output, but also help reduce maintenance costs and extend the service life of equipment. Therefore, accurately evaluating the operating status of wind turbines and promptly discovering and handling possible abnormal situations are crucial to ensuring the economic benefits and technical safety of wind power projects.

[0003] Traditional wind turbine condition assessment mainly relies on regular physical inspections and rule-based monitoring systems. These methods usually rely on setting thresholds and simple statistical models to judge the operating status of the unit. Although these methods are widely used in practice, traditional methods often detect problems only after a fault occurs, making it difficult to perform predictive maintenance. The large amount of collected operating data is not fully utilized, making it difficult to deeply understand the complexity of equipment operation. The assessment process may be limited by the skills and experience of the executors, and subjective judgments affect the accuracy and consistency of the assessment to some extent. Therefore, there is an urgent need for an assessment method and system that can effectively improve the efficiency and accuracy of wind turbine condition assessment. Summary of the invention

[0004] The present disclosure provides a method and system for evaluating the operating status of a wind turbine generator set based on forward-forward propagation, which solves the technical problems of low evaluation efficiency and poor accuracy of wind turbine generator set evaluation methods in the prior art.

[0005] According to a first aspect of the present disclosure, a method for evaluating the operating status of a wind turbine generator system based on forward-forward propagation is provided, comprising the following steps:

[0006] Acquire historical operation data of the wind turbine generator set, draw different operation characteristic curves according to the historical operation data, use a clustering method to process the different operation characteristic curves to obtain an original data set, and divide the original data set into a training set and a test set;

[0007] Generate a label image based on a spatial extension label method, and superimpose it with the training set to obtain positive and negative sample images;

[0008] Inputting the positive and negative sample images into a ResNet network model based on a forward-forward propagation method for training, optimizing the model and parameters using a grid search method, and obtaining a trained network model;

[0009] Real-time operation data is acquired from the wind turbine group and processed, and the processed real-time operation data is input into the network model for prediction and evaluation to obtain the evaluation result of the wind turbine operation status.

[0010] According to the above aspects and any possible implementation, there is further provided an implementation, wherein the historical operation data of the wind turbine generator set includes wind speed data, hub speed data, torque data, blade angle data and power data;

[0011] The operating characteristic curves include a wind speed-power curve, a torque-hub speed curve, a hub speed-blade angle curve, a wind speed-hub speed curve and a wind speed-blade angle curve.

[0012] According to the above aspects and any possible implementation, an implementation is further provided, wherein the process of using a clustering method to process different operating characteristic curves to obtain an original data set and dividing the data set into a training set and a test set is as follows:

[0013] Acquire a standard curve of wind turbine operation characteristics, and calculate a first distance between each data point of the operation characteristic curve and the standard curve of the operation characteristic;

[0014] A threshold is set, and data points corresponding to the first distance being less than or equal to the threshold are regarded as normal data points, and data points corresponding to the first distance being greater than the threshold are regarded as abnormal data points;

[0015] If the ratio of the normal data point to all data points is greater than or equal to a predetermined value, the operation characteristic curve is determined to be normal, otherwise the operation characteristic curve is determined to be abnormal, thereby obtaining a normal curve image and an abnormal curve image;

[0016] The pixels of the normal curve image and the abnormal curve image are adjusted and normalized to obtain an original data set, and the original data set is divided into a training set and a test set in a ratio of 8:2.

[0017] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the process of acquiring the positive and negative sample images is as follows:

[0018] Generate a normal operation label image and an abnormal operation label image of a specific grayscale wave pattern based on a spatial extension label method, wherein the normal operation label image can reflect that the operation condition of the fan is normal, and the abnormal operation label image can reflect that the operation condition of the fan is abnormal;

[0019] Combining the normal curve image and the abnormal curve image with the normal operation label image and the abnormal operation label image respectively to obtain a positive sample image and a negative sample image;

[0020] The pixels of the positive sample image and the negative sample image are uniformly adjusted and normalized to obtain processed positive and negative sample images.

[0021] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the positive sample image includes an image obtained by combining the normal curve image with the normal operation label image and an image obtained by combining the abnormal curve image with the abnormal operation label image;

[0022] The negative sample images include an image obtained by combining the normal curve image with the abnormal operation label image and an image obtained by combining the abnormal curve image with the normal operation label image.

[0023] According to the above aspects and any possible implementation, an implementation is further provided.

[0024] The ResNet network model based on the forward-forward propagation method includes an input layer, a convolutional layer, a maximum pooling layer, a residual block, an average pooling layer and an output layer;

[0025] The convolution layer includes multiple 7×7 convolution kernels; the residual block includes several ResNet50 feature blocks, and the ResNet50 feature block includes several layers of 3×3 convolution layers and ReLU activation functions.

[0026] According to the above aspects and any possible implementation, an implementation is further provided, wherein the process of acquiring real-time operation data from the wind turbine generator set and processing the data is as follows:

[0027] The real-time operation data image is combined with the normal operation label image and the abnormal operation label image respectively to obtain a real-time operation data image with an operation label, wherein the real-time operation data image with an operation label includes a real-time operation data image with a normal operation label and a real-time operation data image with an abnormal operation label.

[0028] According to the above aspects and any possible implementation, an implementation is further provided, wherein the process of inputting the real-time operation data into the network model for prediction and evaluation to obtain the wind turbine operation status evaluation result is:

[0029] Inputting one of the real-time operation data images of the normal operation label or the real-time operation data image of the abnormal operation label into the network model for processing to obtain an operation label classification probability array, wherein the operation label classification probability array includes the operation state of the wind turbine and its corresponding probability value, and the operation state of the wind turbine is normal or abnormal;

[0030] Taking the wind turbine operation state with a large probability value in the operation label classification probability array as the first prediction category;

[0031] Determining whether the first prediction category and the operation label of one of the operation label real-time operation data images point to the same wind turbine operation state;

[0032] If yes, then the first prediction category is used as the wind turbine operation status assessment result;

[0033] If not, another running label real-time running data image is input into the network model for processing, and a second prediction category is determined;

[0034] Determining whether the second prediction category and the operation label of the another operation label real-time operation data image point to the same wind turbine operation state;

[0035] If yes, then the second prediction category is used as the wind turbine operation status assessment result;

[0036] If not, the one with the larger probability value between the first prediction category and the second prediction category is used as the wind turbine operation status evaluation result.

[0037] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the method further includes:

[0038] The ResNet network model based on the forward-forward propagation method is embedded in the SCADA data collection system, the data from the SCADA system is continuously monitored through the model, and the fan operation characteristic curve is drawn, and the fan operation condition is judged according to the fan operation characteristic curve.

[0039] According to a second aspect of the present disclosure, a wind turbine operating status assessment system based on forward-forward propagation is provided, comprising: a wind turbine operating data acquisition module, an operating data processing module, a wind turbine operating status assessment module and a storage module;

[0040] The wind turbine operating data acquisition module is connected to the operating data processing module and is used to collect wind turbine operating data and transmit it to the operating data processing module;

[0041] The operation data processing module is connected to the wind turbine operation status evaluation module, and is used to analyze and process the wind turbine operation data and transmit the processed data to the wind turbine operation status evaluation module;

[0042] The wind turbine operating status evaluation module is used to evaluate the wind turbine operating status according to the processed data;

[0043] The storage module is connected to the wind turbine operation status evaluation module and is used to store the network model.

[0044] The present invention discloses the following technical effects:

[0045] The present invention uses the forward-forward propagation method combined with the residual neural network to train and predict the fan operating characteristic curve, and uses the residual network to refine the seasonality and trend changes of time series data. The powerful feature extraction capability of ResNet can effectively capture the complex patterns in the monthly data, thereby improving the accuracy and robustness of the model for time series anomaly detection and prediction. By introducing the positive and negative sample mechanism of the forward-forward method in the network model, the features and label information in the data can be effectively captured and learned, thereby improving the timeliness and accuracy of fault detection.

[0046] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which:

[0048] Figure 1 A flow chart of a method for evaluating the operating status of a wind turbine generator system based on forward-forward propagation according to an embodiment of the present disclosure is shown;

[0049] Figure 2 A schematic diagram of a normal operation label image and an abnormal operation label image of a wind turbine operating state assessment method based on forward-forward propagation according to an embodiment of the present disclosure is shown;

[0050] Figure 3 A schematic diagram of the structure of a ResNet network model based on a forward-forward propagation method of a wind turbine operating state assessment method based on a forward-forward propagation method according to an embodiment of the present disclosure is shown;

[0051] Figure 4A schematic diagram of positive and negative sample images of wind speed and power of wind turbine A and wind turbine B according to a method for evaluating the operating status of a wind turbine set based on forward-forward propagation according to an embodiment of the present disclosure is shown;

[0052] Figure 5 A schematic structural diagram of a wind turbine operating status assessment system based on forward-forward propagation according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0054] Deep learning technology has made significant progress in image recognition, speech processing and other fields. Its powerful data processing and pattern recognition capabilities are gradually being applied to the operation and maintenance of industrial equipment. Compared with traditional methods, deep learning models can learn from historical data the subtle changes before equipment failures, and achieve failure prediction and prevention. These models can process and analyze large amounts of multi-dimensional data, extract useful features, and improve the accuracy of evaluation. Deep learning can automatically process monitoring data, reduce manual intervention, and improve processing efficiency and consistency.

[0055] Applying deep learning to assess the status of wind turbines can significantly improve the timeliness and accuracy of fault detection. Through continuous learning and optimization, deep learning models can gradually adapt to different working conditions and environmental changes, thereby maintaining efficient monitoring performance under dynamically changing operating conditions. In addition, deep learning can also help operators optimize maintenance strategies and energy management by analyzing the operating data of equipment, further improving the operating efficiency and economic benefits of wind farms. Therefore, using deep learning models to assess the status of wind turbines can not only improve the accuracy and timeliness of fault prediction, but also support smarter decision-making through efficient data analysis.

[0056] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] Please refer to the attached Figure 1 The present invention provides a method for evaluating the operating status of a wind turbine generator system based on forward-forward propagation, comprising the following steps:

[0058] S101. Obtain original operation data of the wind turbine from the SCADA system, draw different operation characteristic curves according to the operation data, use a clustering method to process the different operation characteristic curves to obtain an original data set, and divide the original data set into a training set and a test set.

[0059] This embodiment obtains the operating data of the wind turbine from the SCADA system, including key parameters such as wind speed, hub speed, torque, pitch angle data and power.

[0060] After the data is collected, the operating data is plotted on a monthly basis to obtain different operating characteristic curves, including wind speed-power curve, torque-speed curve, speed-blade angle curve, wind speed-speed curve and wind speed-blade angle curve.

[0061] After drawing the operating characteristic curves, the DBSCAN clustering method is used to divide the wind speed-power curve, torque-speed curve, speed-blade angle curve, wind speed-speed curve and wind speed-blade angle curve into normal curves and abnormal curves according to the characteristics of the curves.

[0062] Specifically, since each fan has its standard operating characteristic curve, the first distance from each data point to the standard operating characteristic curve is calculated. ,

[0063] (1);

[0064] In the formula, Indicates Data points To the point on the standard wind speed-power curve The distance is used to calculate the Euclidean distance between the data point and the standard power curve.

[0065] A threshold is set, and data points corresponding to the first distance being less than or equal to the threshold are regarded as normal data points, and data points corresponding to the first distance being greater than the threshold are regarded as abnormal data points. The threshold range of this embodiment is 20-300, specifically:

[0066] The data points adjacent to the standard operating characteristic curve are defined as a normal cluster, specifically:

[0067] (2);

[0068] (3);

[0069] (4);

[0070] (5);

[0071] In the formula, Indicates Data points To the point on the standard operating characteristic curve The distance is used to calculate the Euclidean distance between the data point and the standard power curve. Indicate point fields, Indicate point and The distance between.

[0072] If the point The domain of has at least minPts points, then the point is a core point. If is the core point, and point exist In the field of from Direct density is reachable if there is a series of points , so that each point is directly density-reachable from the previous point, then from Direct density can be achieved.

[0073] If the ratio of the normal data point to all data points is greater than a predetermined value, the operating characteristic curve is considered normal, otherwise the operating characteristic curve is considered abnormal, and then a normal curve image and an abnormal curve image are obtained. The predetermined value set in this embodiment is 0.7-1; specifically:

[0074] (6);

[0075] Where: is the normal number of data points, is the number of all data points, It is the ratio of normal data points. If it is greater than 0.7, the curve is considered normal, otherwise it is abnormal.

[0076] After obtaining the normal curve image and the abnormal curve image, they are adjusted to a uniform size of 224x224 pixels and normalized so that the pixel values ​​fall between 0 and 1. After image processing, all images are divided into training sets and test sets with a ratio of 8:2.

[0077] The resolution of wind speed and other data in the SCADA system of this embodiment is 10 minutes, and the curves are all scattered point curves.

[0078] S102, generating a label image based on a spatial extension label method, and superimposing it with the training set to obtain positive and negative sample images.

[0079] In order to identify normal curves and abnormal curves, this embodiment sets normal operation labels and abnormal operation labels, such as Figure 2 As shown, the normal and abnormal waveforms generated by this embodiment are vertical waveforms and horizontal waveforms, respectively corresponding to normal operation labels and abnormal operation labels, wherein (a) is a schematic diagram of a normal operation label image, and (b) is a schematic diagram of an abnormal operation label image.

[0080] The specific formula for generating waveform images is as follows:

[0081] (7);

[0082] Where: or Determines the direction of the waveform, is the number of waveforms, and are the width and height of the grayscale wave image, Represents the position in the grayscale wave image The gray value of .

[0083] The grayscale wave image of the normal operation label and the abnormal operation label of this embodiment consists of a grayscale value wave with a fixed frequency, phase and direction, and is the same size as the training data set image. Each possible training set data is associated with a specific configuration of these three parameters. The label information is spatially presented on the entire image, ensuring that each possible filter position can capture the label information.

[0084] After obtaining the normal operation label image and the abnormal operation label image, they are respectively combined with the normal curve and the abnormal curve to obtain the positive sample image and the negative sample image, wherein the positive sample image includes the image obtained by combining the normal curve image with the normal operation label image and the image obtained by combining the abnormal curve image with the abnormal operation label image; the negative sample image includes the image obtained by combining the normal curve image with the abnormal operation label image and the image obtained by combining the abnormal curve image with the positive label. The sample generation formula is as follows:

[0085] (8);

[0086] (9);

[0087] Where: is a positive sample image, is a negative sample image, is the normal curve image, is the abnormal curve image, To properly run the label image, Label images for abnormal runs.

[0088] After the positive and negative sample image data are obtained by combining, the positive and negative samples are unified into the ResNet50 data format and the images are resized to 224x224 pixels. The pixel values ​​of the resized images are normalized to the range of [0, 1] using the mean and standard deviation of ImageNet. The formula is as follows:

[0089] (10);

[0090] (11);

[0091] (12);

[0092] Where: , .

[0093] The spatial extension labeling method of this embodiment can enhance the expression of data labels, so that the deep learning model can more effectively learn features related to working condition labels from images.

[0094] S103, inputting the positive and negative sample images into a ResNet network (FF-ResNet) model based on a forward-forward propagation method for training, and optimizing the model and parameters using a grid search method.

[0095] The FF-ResNet model structure of this embodiment is as follows Figure 3 As shown, the model has a total of five layers: input layer, convolutional layer, residual block, average pooling layer and output layer.

[0096] The first layer is the input layer, which mainly receives image data as input. The image data first enters the network through the input layer;

[0097] The second layer is the convolution layer, which includes a 7x7 initial convolution layer and subsequent multiple 3x3 convolution layers. It uses multiple convolution kernels of different sizes to slide on the input data to extract basic visual features such as edges and textures. The specific formula is as follows:

[0098] (13);

[0099] In the formula, is the input feature map, is the convolution kernel, is the bias, is the output feature map, is the size of the convolution kernel, is the number of channels of the input feature map, is the channel index of the output feature map.

[0100] The third layer is the maximum pooling layer, which is used to reduce the spatial dimension of the feature map, thereby reducing computational complexity and memory requirements while retaining the most significant feature information. The specific formula is as follows:

[0101] (14);

[0102] In the formula, is the position in the output feature map The value of is the position in the input feature map The value of Indicates the position of the input feature map and the output feature map The corresponding window area. Indicates in the window The maximum value found in .

[0103] The fourth layer is the residual block, which includes multiple ResNet50 feature blocks, preferably 4. Each feature block contains several layers of 3x3 convolutional layers, all of which apply the ReLU activation function. Feature extraction is performed inside each residual block and the extracted output is added to the input feature through the jump connection, which helps to solve the gradient disappearance problem in deep networks. The specific formula is as follows:

[0104] (15);

[0105] In the formula, are input features, represents the transformation after several convolutional layers, is the convolution kernel.

[0106] The fifth layer is the average pooling layer. After the deep convolution, the average pooling layer is used to reduce the feature dimension and help reduce the overfitting of the model, reduce the amount of calculation, and retain important feature information. The specific formula is as follows:

[0107] (16);

[0108] Where: and is the height and width of the input feature map, is the channel index.

[0109] The sixth layer is the output layer, which usually includes a fully connected layer and an activation function to map the extracted features to the final output category.

[0110] After training, a loss function is constructed based on the output results. The loss function of the ResNet network (FF-ResNet) model based on the forward-forward propagation method in this embodiment is a logarithmic loss function combined with a threshold mechanism. The network weights are manually adjusted by two forward propagations of positive and negative samples. The combination of positive and negative samples enables the loss function to have different sensitivities to positive sample data and negative sample data during training. By quantifying the difference between positive and negative samples, the model is helped to learn and distinguish the characteristics of normal data and abnormal data more accurately.

[0111] The resulting loss function is as follows:

[0112] (17);

[0113] Where: is the number of samples, is the score of the ith positive sample, is the score of the i-th negative sample, It is the threshold of the loss function, which is used to adjust the classification boundary of the model.

[0114] The loss function of the ResNet network (FF-ResNet) model based on the forward-forward propagation method in this embodiment is a logarithmic loss function. The goal of the loss function is to minimize the error. It combines the positive sample prediction score and the negative sample prediction score, and uses a threshold to increase the accuracy of the positive class score while reducing the impact of the negative class score.

[0115] The neural network weights and bias parameters of this embodiment enable the model to perform effective prediction or classification by learning the complex relationship between input data and output labels. Through complex network structures and loss functions, efficient and accurate classification of image data is achieved. The model optimizes network weights and bias parameters by learning the complex relationship between input data and output labels, thereby improving the accuracy of prediction and the generalization ability of the model.

[0116] S104, acquiring real-time operation data from the wind turbine and processing it, inputting the processed real-time operation data into a ResNet network (FF-ResNet) model based on a forward-forward propagation method for prediction, obtaining a prediction result, and evaluating the operation status of the wind turbine according to the prediction result.

[0117] This embodiment obtains and processes real-time operation data from a wind turbine. Specifically, this embodiment combines the normal operation label image and the abnormal operation label image with the operation characteristic curve in the real-time operation data, and uses all images and all labels in the real-time operation data to generate feature-label data, so that each image will generate a set of feature data that combines normal label information and abnormal label information, and obtains real-time operation data that contains normal operation label images and abnormal operation label images. For each test image Generate two sets of test data. The specific formula is as follows:

[0118] (18);

[0119] (19);

[0120] In the formula, It is real-time running data combined with a waveform image representing a normal label. It is real-time operation data combined with waveform images representing abnormal labels. It is a real-time running data image. and There are two types of labeled waveform images: normal and abnormal.

[0121] Input the real-time running data into the trained prediction model for prediction. The model will output a set of results, that is, a set of classification probability arrays. For each probability array, select the sequence number with the largest probability and the index of the maximum value in the array as the prediction category. The formula is as follows:

[0122] (20);

[0123] (twenty one);

[0124] (twenty two);

[0125] (twenty three);

[0126] In the formula, and They are the classification probability arrays of real-time running data with normal operation labels and real-time running data with abnormal operation labels, and They are the classification categories of normal operation labels and the classification categories of abnormal operation labels.

[0127] Specifically, this implementation inputs either the real-time operation data of the normal operation label or the real-time operation data of the abnormal operation label into the network model for processing to obtain an operation label classification probability array, wherein the operation label classification probability array includes the fan operation status and its corresponding probability value, and the fan operation status is normal or abnormal.

[0128] Then, the wind turbine operating state with a large probability value in the operating label classification probability array is taken as the first prediction category; it is determined whether the first prediction category and the operating label of the operating label real-time operating data point to the same wind turbine operating state;

[0129] If yes, the first prediction category is used as the wind turbine operation status evaluation result;

[0130] If not, another type of running label real-time running data is input into the network model for processing, and a second prediction category is determined in the same way;

[0131] Determine whether the second prediction category and the operation tag of another operation tag real-time operation data point to the same wind turbine operation state;

[0132] If yes, the second prediction category is used as the wind turbine operation status evaluation result;

[0133] If not, the one with the larger probability value in the first prediction category and the second prediction category is taken as the evaluation result of the wind turbine operation status.

[0134] The specific formula is as follows:

[0135] (twenty four);

[0136] In the formula, is a normal label image or an abnormal label image, is the final evaluation result of the prediction.

[0137] S105. The model is embedded in the SCADA data collection software. In daily operations, the model continuously monitors new data from the SCADA system and draws the latest wind turbine operating characteristic curve. Once an abnormal operating condition is detected, the system automatically collects this information. It can not only help the operation and maintenance team respond to potential problems in a timely manner, but also further optimize the performance of the model through continuous real-time data feedback.

[0138] According to the above-mentioned embodiments of the present disclosure, the forward-forward propagation method is combined with the residual neural network to train and predict the wind turbine operating characteristic curve, and the residual network is used to refine the seasonality and trend changes of the time series data. The powerful feature extraction capability of ResNet can effectively capture the complex patterns in the monthly data, thereby improving the accuracy and robustness of the model for time series anomaly detection and prediction. By introducing the positive and negative sample mechanism of the forward-forward method in the network model, the features and label information in the data can be effectively captured and learned, thereby improving the timeliness and accuracy of fault detection.

[0139] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present disclosure is not limited by the order of the actions described, because according to the present disclosure, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present disclosure.

[0140] The above is an introduction to the method embodiment. The following is a further explanation of the scheme disclosed in the present invention through a system embodiment.

[0141] like Figure 5 As shown, this embodiment also provides a wind turbine operating status assessment system based on forward-forward propagation, including: a wind turbine operating data acquisition module 101, an operating data processing module 102, a wind turbine operating status assessment module 103 and a storage module 104;

[0142] The wind turbine operation data collection module 101 is connected to the operation data processing module 102 and is used to collect the wind turbine operation data and transmit it to the operation data processing module;

[0143] The operation data processing module 102 is connected to the wind turbine operation status evaluation module 103, and is used to analyze and process the wind turbine operation data and transmit the processed data to the wind turbine operation status evaluation module;

[0144] The wind turbine operating status evaluation module 103 is used to evaluate the wind turbine operating status according to the processed data;

[0145] The storage module 104 is connected to the wind turbine operation status evaluation module 103 and is used to store the network model and continuously monitor new data from the SCADA system.

[0146] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0147] Embodiment: This embodiment takes the wind speed-power operating characteristic curve as an example, and selects the wind speed and power data of two wind turbines A and B from September 1 to 30, 2019 in the SCADA system, as shown in Table 1:

[0148] Table 1 Wind speed and power data of the fan collected in this embodiment

[0149]

[0150] After the data collection is completed, the wind speed-power operating characteristic curves of wind turbine A and wind turbine B are obtained with wind speed as the horizontal axis and power as the vertical axis.

[0151] Then, the wind turbine wind speed-power operation standard characteristic curve and threshold are obtained, and the first distance from each data point to the wind speed-power operation characteristic curve is calculated. The threshold and the first distance are compared to obtain normal data points and abnormal data points. In this embodiment, the threshold is preferably 200.

[0152] After obtaining the normal data points, the ratio of the normal data points to all the data points is calculated. When the ratio is greater than 0.7, the curve is considered to be a normal curve. Among them, the wind speed-power curve of wind turbine A in September 2019 is an abnormal curve, and the wind speed-power curve of wind turbine B in September 2019 is a normal curve.

[0153] Generate a normal operation label image and an abnormal operation label image, and combine them with the normal curve and the abnormal curve to obtain a positive sample image and a negative sample image. The positive sample image and the negative sample image obtained in this embodiment are as follows: Figure 4 As shown, (a) is the positive sample image obtained by combining the operating characteristic curve of fan A with the normal operating label image, (b) is the negative sample image obtained by combining the operating characteristic curve of fan A with the abnormal operating label image, (c) is the negative sample image obtained by combining the operating characteristic curve of fan B with the normal operating label image, and (d) is the positive sample image obtained by combining the operating characteristic curve of fan B with the abnormal operating label image.

[0154] After the model is trained, a test data set is constructed, and the wind speed-power operation characteristic curves of wind turbine A and wind turbine B are respectively combined with the normal operation label image and the abnormal operation label image to obtain the test set data containing the normal operation label images and the abnormal operation label images of wind turbine A and wind turbine B.

[0155] The above data is input into the model for prediction, and the resulting running label classification probability array is shown in Table 2:

[0156] Table 2 Model prediction results of fan A and fan B obtained in this embodiment

[0157]

[0158] The prediction category is determined according to the above-mentioned operation label classification probability array, wherein the abnormal probability in the data of fan A combined with the normal label is greater than the normal probability, then the abnormality is taken as the prediction category. Since the prediction category and the normal operation label do not point to the same fan operation status, the information combined with the abnormal label is judged. In the data combined with the abnormal label, the prediction category is determined to be abnormal through the above-mentioned method. Since the prediction category and the abnormal operation label point to the same fan operation status, the output prediction result is abnormal. Therefore, it is judged that the operation status of fan A in September 2019 is abnormal.

[0159] At the same time, the normal probability in the data of wind turbine B combined with the normal label is greater than the abnormal probability, so normal is used as the prediction category. Since the prediction category and the normal operation label point to the same wind turbine operating status, a normal result is output. Therefore, it is judged that the operating status of wind turbine B in September 2019 is normal.

[0160] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present disclosure is not limited by the order of the actions described, because according to the present disclosure, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present disclosure.

[0161] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0162] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A wind turbine operating status assessment method based on forward-forward propagation, characterized in that: The following steps are involved: Acquire historical operation data of the wind turbine generator set, draw different operation characteristic curves according to the historical operation data, use a clustering method to process the different operation characteristic curves to obtain an original data set, and divide the original data set into a training set and a test set; The historical operation data of the wind turbine generator set includes wind speed data, hub speed data, torque data, blade angle data and power data; The operating characteristic curves include a wind speed-power curve, a torque-hub speed curve, a hub speed-blade angle curve, a wind speed-hub speed curve and a wind speed-blade angle curve; The process of using a clustering method to process different operating characteristic curves to obtain an original data set and divide it into a training set and a test set is as follows: Obtaining a standard curve of wind turbine operation characteristics, and calculating a first distance between each data point of the operation characteristic curve and the standard curve of the operation characteristics; A threshold is set, and data points corresponding to the first distance being less than or equal to the threshold are regarded as normal data points, and data points corresponding to the first distance being greater than the threshold are regarded as abnormal data points; If the ratio of the normal data point to all data points is greater than or equal to a predetermined value, the operating characteristic curve is determined to be normal, otherwise the operating characteristic curve is determined to be abnormal, thereby obtaining a normal curve image and an abnormal curve image; The pixels of the normal curve image and the abnormal curve image are adjusted and normalized to obtain an original data set, and the original data set is divided into a training set and a test set in a ratio of 8:2; Generate a label image based on a spatial extension label method, and superimpose it with the training set to obtain positive and negative sample images; The process of obtaining the positive and negative sample images is as follows: Generate a normal operation label image and an abnormal operation label image of a specific grayscale wave pattern based on a spatial extension label method, wherein the normal operation label image can reflect that the operation condition of the fan is normal, and the abnormal operation label image can reflect that the operation condition of the fan is abnormal; Combining the normal curve image and the abnormal curve image with the normal operation label image and the abnormal operation label image respectively to obtain a positive sample image and a negative sample image; The pixels of the positive sample image and the negative sample image are uniformly adjusted and normalized to obtain processed positive and negative sample images; The positive sample image includes an image obtained by combining the normal curve image with the normal operation label image and an image obtained by combining the abnormal curve image with the abnormal operation label image; The negative sample images include an image obtained by combining the normal curve image with the abnormal operation label image and an image obtained by combining the abnormal curve image with the normal operation label image; Inputting the positive and negative sample images into a ResNet network model based on a forward-forward propagation method for training, optimizing the model and parameters using a grid search method, and obtaining a trained network model; Real-time operation data is acquired from the wind turbine group and processed, and the processed real-time operation data is input into the network model for prediction and evaluation to obtain the evaluation result of the wind turbine operation status.

2. The wind turbine operating status assessment method based on forward-forward propagation according to claim 1 is characterized in that: The ResNet network model based on the forward-forward propagation method includes an input layer, a convolutional layer, a maximum pooling layer, a residual block, an average pooling layer and an output layer; The convolution layer includes multiple 7×7 convolution kernels; the residual block includes several ResNet50 feature blocks, and the ResNet50 feature block includes several layers of 3×3 convolution layers and ReLU activation functions.

3. The wind turbine operating status assessment method based on forward-forward propagation according to claim 1 is characterized in that: The process of obtaining real-time operation data from the wind turbine generator set and processing it is as follows: The real-time operation data image is combined with the normal operation label image and the abnormal operation label image respectively to obtain a real-time operation data image with an operation label, wherein the real-time operation data image with an operation label includes a real-time operation data image with a normal operation label and a real-time operation data image with an abnormal operation label.

4. The method for evaluating the operating status of a wind turbine generator system based on forward-forward propagation according to claim 3 is characterized in that: The process of inputting the real-time operation data into the network model for prediction and evaluation to obtain the wind turbine operation status evaluation result is as follows: Inputting one of the real-time operation data images with normal operation labels or the real-time operation data images with abnormal operation labels into the network model for processing to obtain an operation label classification probability array, wherein the operation label classification probability array includes the operation status of the fan and its corresponding probability value, and the operation status of the fan is normal or abnormal; Taking the wind turbine operation state with a large probability value in the operation label classification probability array as the first prediction category; Determining whether the first prediction category and the operation label of one of the operation label real-time operation data images point to the same wind turbine operation state; If yes, then the first prediction category is used as the wind turbine operation status assessment result; If not, another running label real-time running data image is input into the network model for processing, and a second prediction category is determined; Determining whether the second prediction category and the operation label of the another operation label real-time operation data image point to the same wind turbine operation state; If yes, then the second prediction category is used as the wind turbine operation status assessment result; If not, the one with the larger probability value between the first prediction category and the second prediction category is used as the wind turbine operation status evaluation result.

5. The method for evaluating the operating status of a wind turbine generator system based on forward-forward propagation according to claim 1, characterized in that: The method further comprises: The ResNet network model based on the forward-forward propagation method is embedded in the SCADA data collection system, the data from the SCADA system is continuously monitored through the model, and the fan operation characteristic curve is drawn, and the fan operation condition is judged according to the fan operation characteristic curve.

6. A wind turbine operating status assessment system based on forward-forward propagation, implemented by the method according to any one of claims 1 to 5, characterized in that: include: A wind turbine operating data acquisition module (101), an operating data processing module (102), a wind turbine operating status evaluation module (103) and a storage module (104); The wind turbine operating data collection module (101) is connected to the operating data processing module (102) and is used to collect wind turbine operating data and transmit the data to the operating data processing module; The operation data processing module (102) is connected to the wind turbine operation status evaluation module (103) and is used to analyze and process the wind turbine operation data and transmit the processed data to the wind turbine operation status evaluation module; The wind turbine operating status evaluation module (103) is used to evaluate the wind turbine operating status according to the processed data; The storage module (104) is connected to the wind turbine operation status assessment module (103) and is used to store the network model.

Citation Information

Patent Citations

  • Fault diagnosis method for planetary gearbox of wind turbine generator

    CN113505664A

  • Fault early warning system of new energy fan equipment

    CN115982663A

  • Comprehensive evaluation method for overall performance of wind turbine generator under normal and extreme working conditions

    CN117333029A

  • Intelligent monitoring method for offshore wind turbine generator based on perception calculation collaboration

    CN117989074A

  • Method and system for condition monitoring of wind turbines

    KR101345598B1