Wind turbine generator data processing methods, systems, electronic equipment and storage media

Abstract

This application relates to a data processing method, system, electronic device, and storage medium for wind turbine generator sets. The method includes: real-time acquisition and uploading of multi-dimensional operational data from the wind turbine generator set; preprocessing and normalizing the multi-dimensional operational data, and extracting key features from the processed data using a convolutional neural network to obtain key feature data; acquiring historical datasets and a Logistic regression model, training the Logistic regression model based on the historical dataset to obtain the fault early warning model; calculating the probability of fault occurrence using the fault early warning model on the key feature data, and issuing an early warning signal when the probability of fault occurrence exceeds a preset threshold. Through precise data processing and model training, higher model fit and smaller prediction errors are ensured, enabling more accurate prediction of faults in the wind turbine generator set's drivetrain, achieving early warning, and improving the accuracy and reliability of fault early warning.

Description

Technical Field

[0001] This application relates to the field of power data processing, and in particular to methods, systems, electronic devices and storage media for processing data from wind turbine generator sets. Background Technology

[0002] Wind turbine generators may encounter various faults during operation, among which drivetrain failure is one of the most common. Traditional fault diagnosis methods usually rely on human experience and offline monitoring, which have problems such as low diagnostic accuracy and slow response speed. As the core power generation component of wind turbine generators, the failure of the wind turbine drivetrain will bring huge maintenance and replacement costs.

[0003] Therefore, there is an urgent need for a new data processing solution for wind turbine generators to solve the above-mentioned technical problems. Summary of the Invention

[0004] This application provides a data processing method, system, electronic device, and storage medium for wind turbine generator sets, which at least solves the problems of low diagnostic accuracy and slow response speed in traditional fault diagnosis methods in related technologies.

[0005] In a first aspect, embodiments of this application provide a method for processing data from a wind turbine generator set, including:

[0006] Real-time acquisition and uploading of multi-dimensional operational data from wind turbine generators;

[0007] The multidimensional operational data is preprocessed and normalized, and key features are extracted from the processed data through a convolutional neural network to obtain key feature data.

[0008] Obtain a historical dataset and a Logistic regression model, train the Logistic regression model based on the historical dataset, and obtain the fault warning model. The historical dataset includes multi-dimensional operational data collected in the past.

[0009] The fault early warning model calculates the probability of fault occurrence by analyzing the key feature data. When the probability of fault occurrence is greater than a preset threshold, an early warning signal is issued.

[0010] In one embodiment, the preprocessing and normalization of the multidimensional operational data includes:

[0011] The multidimensional operational data is decompressed, and missing values ​​are imputed in the decompressed multidimensional operational data according to the physical characteristics of different measurement points to obtain complete multidimensional operational data.

[0012] The quartile method is used to calculate the quartiles in the complete multidimensional running data. The upper and lower limits of outliers in the data are determined based on the quartiles, and the complete multidimensional running data that is not within the upper and lower limits is removed to obtain normal multidimensional running data.

[0013] The normalized multidimensional operating data is then processed to obtain the data to be characterized.

[0014] In one embodiment, the step of calculating the quartiles in the complete multidimensional operational data using the quartile method, determining the upper and lower limits of outliers in the data based on the quartiles, and processing the complete multidimensional operational data that is not within the upper and lower limits to obtain normal multidimensional operational data includes:

[0015] Calculate the first quartile and the third quartile of the multidimensional operational data;

[0016] Calculate the interquartile range based on the first quartile and the third quartile;

[0017] The upper and lower limits of outliers are determined based on the interquartile range.

[0018] Identify and mark outliers from the multidimensional operational data based on the upper and lower limits;

[0019] The system handles outlier values ​​by deleting, replacing, or formatting them.

[0020] In one embodiment, the step of extracting key features from the preprocessed data using a convolutional neural network to obtain key feature data includes:

[0021] Obtain a pre-built convolutional neural network architecture, which includes an input layer, convolutional layers, pooling layers, and fully connected layers;

[0022] The system receives data to be characterized through an input layer, extracts specific features from the data through a convolutional layer to obtain a key feature matrix, performs dimensionality reduction on the key feature matrix through a pooling layer, and maps the dimensionality-reduced key feature matrix to the final output category through a fully connected layer to obtain the key feature data.

[0023] In one embodiment, the construction of the fault early warning model includes:

[0024] Obtain the risk factors and corresponding risk factor coefficients for the occurrence of failures, and construct an initial failure early warning model based on the risk factors, risk factor coefficients, and Logistic regression model;

[0025] The coefficients of the initial fault warning model are solved by the maximum likelihood estimation method to obtain the regression coefficients. The regression coefficients are then substituted into the initial fault warning model to obtain the fault warning model.

[0026] In one embodiment, the method further includes:

[0027] Select the appropriate SVM kernel function according to user needs, and use grid search and cross-validation methods to adjust the parameters of the SVM to obtain the initial SVM model;

[0028] Obtain a labeled training dataset to train the SVM model, and use a test dataset to evaluate the performance of the SVM model.

[0029] The trained SVM model is used to predict the fault category of the key feature data, and a nonlinear kernel function is used for optimal calculation to obtain the fault attribute classification of different key feature data.

[0030] Secondly, embodiments of this application provide a wind turbine generator data processing system for running the wind turbine generator data processing method described in any of the above embodiments, the system comprising:

[0031] The data acquisition module is used to collect and upload multi-dimensional operating data of the wind turbine generator in real time, and send the multi-dimensional operating data to the data processing module.

[0032] The data processing module is used to preprocess and normalize the multidimensional operating data, and extract key features from the preprocessed data through a convolutional neural network to obtain key feature data and send it to the fault warning module.

[0033] The data iteration module is used to acquire historical datasets and a Logistic regression model, train the Logistic regression model based on the historical datasets, obtain the fault warning model, and send it to the fault warning module. The historical datasets include multi-dimensional operational data collected in the past.

[0034] The fault early warning module is used to receive the key feature data and run the fault early warning model. The fault early warning model is used to calculate the key feature data to obtain the probability of fault occurrence. When the probability of fault occurrence is greater than a preset threshold, an early warning signal is issued.

[0035] In one embodiment, the data acquisition module includes a buffer unit, which is communicatively connected to multiple acquisition points of the wind turbine generator set;

[0036] The real-time acquisition and uploading of multi-dimensional operating data of the wind turbine generator set, and the sending of the multi-dimensional operating data to the data processing module, includes:

[0037] The cache unit is monitored in real time. When new collected data is detected to be recorded, the new multidimensional running data is compressed and encrypted to obtain multidimensional running data.

[0038] The data acquisition module uses a low-latency transmission protocol and multiplexing to transmit the multidimensional operational data from the acquisition point to the data processing module.

[0039] An anomaly detection mechanism is used to monitor the data transmission process. If a data transmission failure is detected, the multidimensional operational data is resent.

[0040] In one embodiment, the data processing module includes a preprocessing unit and a feature extraction unit, wherein,

[0041] The preprocessing unit is used to preprocess and normalize the multidimensional operational data;

[0042] The feature extraction unit is used to extract key features from the preprocessed data using a convolutional neural network.

[0043] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the wind turbine generator data processing method as described in the first aspect above.

[0044] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the wind turbine generator data processing method as described in the first aspect above.

[0045] The wind turbine generator data processing method, system, electronic equipment, and storage medium provided in this application have at least the following technical effects:

[0046] Anomaly processing using the quartile method ensures data quality and reliability. Data normalization ensures features are within the same scale, facilitating subsequent feature extraction and model training. Convolutional neural networks extract key features from the processed data, ensuring a high correlation between the extracted features and fault warnings. Support vector machine algorithms and nonlinear kernel functions are used to classify fault warning data attributes, improving fault identification accuracy. A logistic regression algorithm is employed to implement the fault warning model, predicting the probability of fault occurrence. Maximum likelihood estimation is used to solve for regression coefficients, ensuring optimal estimation of model parameters and improving the accuracy and reliability of fault warnings. Precise data processing and model training ensure higher model fit and smaller prediction errors, enabling more accurate prediction of wind turbine drivetrain faults and achieving early warning.

[0047] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0048] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0049] Figure 1 This is a structural block diagram of a wind turbine generator data processing system according to one embodiment of this application;

[0050] Figure 2 This is a flowchart of a wind turbine generator data processing method according to an embodiment of this application;

[0051] Figure 3 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0053] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0054] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0055] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0056] refer to Figure 1This application provides a wind turbine generator data processing system. The system includes a data acquisition module, a data processing module, a fault early warning module, and a data iteration module. This data processing system is capable of operating... Figure 2 Data processing methods for wind turbine generator sets. The following content combines... Figure 1 and Figure 2 The processing system and processing method of this application are described.

[0057] In this embodiment, the data acquisition module is used to collect and upload multi-dimensional operating data of the wind turbine generator in real time, and send the multi-dimensional operating data to the data processing module; the data processing module is used to preprocess and normalize the multi-dimensional operating data, and extract key features from the preprocessed data through a convolutional neural network to obtain key feature data and send it to the fault warning module; the data iteration module is used to acquire historical datasets and a Logistic regression model, train the Logistic regression model based on the historical datasets to obtain the fault warning model and send it to the fault warning module, wherein the historical datasets include historically collected multi-dimensional operating data; the fault warning module is used to receive the key feature data and run the fault warning model, calculate the fault occurrence probability through the fault warning model, and issue a warning signal when the fault occurrence probability is greater than a preset threshold.

[0058] In this embodiment, the data acquisition module includes a cache unit, which is communicatively connected to multiple acquisition points of the wind turbine generator set. During data acquisition, the cache unit is monitored in real time. When new acquired data is detected, the new multidimensional operational data is compressed and encrypted to obtain multidimensional operational data. The data acquisition module uses a low-latency transmission protocol and multiplexing to transmit the multidimensional operational data from the acquisition points to the data processing module. An anomaly detection mechanism monitors the data transmission process; if a data transmission failure is detected, the multidimensional operational data is resent.

[0059] Specifically, the data acquisition module is not only responsible for collecting the operating data of the wind turbine generator set, including but not limited to multi-dimensional data such as wind speed, temperature, and vibration signals, but also has the ability to transmit the collected data to the edge computing device (i.e., the data processing module) in real time.

[0060] Specifically, the data acquisition module achieves this function in the following ways:

[0061] One is real-time data streaming. The data acquisition module is designed to transmit the acquired data to the edge computing device in real time. Whenever a new data point is recorded, the data will be sent out immediately, instead of waiting for the data to accumulate to a certain amount before being transmitted in batches. This method ensures that the latest data can be received and processed by the edge computing device in a timely manner, thereby improving the timeliness of fault warning.

[0062] Secondly, there is the low-latency transmission protocol. In order to ensure low latency in data transmission, the data acquisition module adopts a transmission protocol that can quickly transmit data from the acquisition point to the edge computing node, reducing data lag caused by network latency. This is especially important for fault early warning systems that require real-time monitoring and rapid response.

[0063] Thirdly, data compression and encryption. Before data transmission, the data acquisition module will compress the data to reduce the time and bandwidth required for transmission. At the same time, in order to protect the security of the data, the data in transmission will also be encrypted to prevent the data from being stolen or tampered with during transmission.

[0064] Fourthly, it features adaptive transmission rate. The data acquisition module has the ability to dynamically adjust the transmission rate. Based on network conditions and the processing power of edge computing devices, it can automatically adjust the data transmission speed to ensure the stability and reliability of data transmission even under poor network conditions.

[0065] Fifthly, there is an anomaly detection and retransmission mechanism. In order to deal with possible packet loss or transmission errors during data transmission, the data acquisition module is also equipped with an anomaly detection and data retransmission mechanism. Once a data transmission failure is detected, the system will automatically resend the relevant data until it is confirmed that the edge computing device has successfully received it.

[0066] Sixthly, it features multiplexing technology. The data acquisition module supports multiplexing technology, which can simultaneously handle data acquisition and transmission tasks from multiple data sources. This allows for efficient centralized data transmission even when dealing with large-scale or multi-point distributed wind turbine generators.

[0067] Through the above measures, the data acquisition module can not only ensure the accuracy and completeness of data acquisition, but also ensure that this data can be transmitted to the edge computing device in a timely and secure manner, providing a solid foundation for subsequent data processing and fault early warning.

[0068] In one detailed embodiment, the preprocessing unit receives data from the data acquisition module, cleans the acquired data, performs outlier processing using the quartile method, determines the upper and lower limits of outliers in the sample data based on the interquartile range, and normalizes the data. To achieve the above functions, the data processing module includes a preprocessing unit and a feature extraction unit. The preprocessing unit preprocesses and normalizes the multidimensional running data; the feature extraction unit extracts key features from the preprocessed data using a convolutional neural network.

[0069] Specifically, the preprocessing unit can clean up dirty and abnormal data in the operating data, and reasonably impute or remove null values ​​in the data according to the physical characteristics of different measuring points. Then, it calculates the reasonable range of data for each measuring point by combining statistical principles and cleans up data that exceeds the range. The high-quality health dataset extracted through the monitoring and control software (SCADA) data preprocessing work is the foundation for building the wind turbine fault early warning model. Using SCADA abnormal data processing methods, abnormal data caused by the uncertainty of wind speed and wind direction are removed. Then, using SCADA data parameter selection methods, information redundancy in SCADA data is removed to avoid the model training getting stuck in local minima or even gradient explosion, which directly affects the model training process. Finally, data normalization is used to eliminate the influence of different dimensions between wind turbine parameters, so as to accelerate the gradient descent speed during model training, improve the prediction accuracy of the model, and potentially extend the prediction step size. In order to eliminate the influence of the corresponding dimensions of various parameters, prevent parameters with large orders of magnitude from dominating during training, and reduce the difference in orders of magnitude to accelerate the iteration convergence speed, the sample data is normalized.

[0070] In the entire data processing and analysis process of wind turbine generator drivetrain fault detection, an artificial intelligence algorithm is designed to adapt to different operating conditions of the wind turbine and achieve the matching relationship between characteristics and fault modes under different operating conditions, so as to achieve better processing results.

[0071] The feature extraction unit uses a convolutional neural network to extract key features from the processed data and implements fault warning data attribute classification based on the support vector machine algorithm and nonlinear kernel function.

[0072] The original data of wind turbine drivetrain failures has multiple dimensions, including redundant and low-correlation feature dimensions. Convolutional neural networks can improve the feature extraction capability of the data through convolutional pooling. Multiple convolutional feature maps can be obtained by using the convolution operation of the convolution kernel, which can effectively extract the key features of the original input data. The convolution operation is expressed as formula (1):

[0073]

[0074] In the formula, x represents the n′-th weight in the m-th convolutional kernel of layer l; l(n+n′) represents the n′ weight-sensing position in the nth convolutional region of the l-th layer; c represents the size of the convolution kernel.

[0075] Feature compression through pooling layers can reduce the dimensionality of the feature matrix, thereby significantly reducing the number of model training parameters and obtaining the main features. Specifically, max pooling takes the maximum value in the receptive region as the output of the pooling layer, as shown below:

[0076] p l(m，t) =y l(m，n) max (n-1)c+1≤1≤jw {a l(m，t)} (2)

[0077] In the formula, a l(m，t) It represents the t-th activation value of the m-th feature map in the l-th layer, where w is the pooling width.

[0078] The above information lays the foundation for early warning of transmission chain faults in wind turbine generators.

[0079] For fault warning data attribute classification based on support vector machine (SVM) algorithm and nonlinear kernel function, the main purpose of wind turbine generator drivetrain fault warning and control is to achieve autonomous and intelligent control of the wind turbine generator drivetrain, reliably analyze the operating status of the wind turbine generator drivetrain, and send fault warnings in advance before abnormalities occur in the drivetrain and cause failures, thereby reducing the risk of drivetrain operation and ensuring multi-dimensional fault warnings. Therefore, the application layer establishes a fault warning model based on the drivetrain operation data processing and analysis of other layers to realize multi-dimensional fault warning and control of the drivetrain. The SVM algorithm is used to establish the fault warning mechanism. Since fault warning and control problems have significant inherent coupling correlation, collaborative requirements, and multi-dimensional changes, the SVM algorithm can deeply mine the data to realize multi-dimensional fault warning and control patterns. Furthermore, a nonlinear kernel function is introduced to perform optimal calculation and realize fault warning data attribute classification. The algorithm is used to perform attribute division on the multi-dimensional data of drivetrain fault warning and control, represented as follows:

[0080]

[0081] In the formula, f(x) represents the optimal classification decision function for the data, the sign function is denoted by sign, b represents the set parameters, x represents the unknown quantity, and α m y represents the regularization coefficient. m Let σ represent a binary variable, and let σ represent the volatility coefficient.

[0082] After completing the data cleaning and preprocessing of the wind turbine generator drivetrain, the data feature extraction based on convolutional neural networks, and the classification of drivetrain early warning data attributes, a wind turbine generator drivetrain fault early warning based on an edge AI computing architecture can be realized.

[0083] Edge computing utilizes chips for high-frequency sampling and real-time statistical calculations, periodically calculating a set of characteristic parameters and transmitting them to a cloud server via a communication module for diagnostic and predictive analysis. The edge AI computing architecture randomly arranges the specific times of data upload within each preset time period to even out the amount of data uploaded as much as possible. Gateways and controllers are considered edge computing nodes, possessing computing, storage, and data transmission capabilities. For wind turbine drivetrain data acquisition, the quartile method is used for outlier processing, dividing the data into four equal parts, where the interquartile range (I0) is... QR It is the difference between the third quartile and the first quartile, expressed as:

[0084] I QR =Q3-Q1 (4)

[0085] In the formula, Q3 represents the third quartile, and Q1 represents the first quartile. When outliers exist in the sample data, the upper and lower limits of the outlier range in the sample data can be determined based on the interquartile range [F]. d， F u The calculation is expressed as follows:

[0086]

[0087] The fault early warning module employs a Logistic regression algorithm to ensure reliable fault early warning. Let E represent the early warning value, which is the threshold for triggering a normal early warning; this application does not impose any range restrictions on E. Many factors can cause risks in the transmission chain, such as temperature and load rate. The Logistic regression fault early warning model is constructed as follows:

[0088]

[0089] In the formula, the regression coefficients of the risk influencing factors are denoted by β1, β2, ..., β3. j Represents: k1, ..., k j Represents the risk factor; p represents the risk probability. After obtaining the transmission chain fault warning result, the probability of the fault occurring is calculated:

[0090]

[0091] T is mainly used to compare the difference between the average values ​​of two sets of data. It is an existing technology for solving regression coefficients using the maximum likelihood estimation method. After obtaining the result from formula (6), it can be substituted into formula (7) to solve formula (7). By using the maximum likelihood estimation method to solve the regression coefficient e and completing the above formula, the probability of fault occurrence can be obtained, and fault identification can be completed. At the same time, an early warning is sent to the display interface to complete the transmission chain fault early warning. Thus, based on the above relevant content, the design of the wind turbine generator transmission chain fault early warning algorithm based on the edge AI computing architecture is completed.

[0092] The data iteration module uses the data classified by the data processing module to build a historical dataset and replaces the initial data one by one. Based on the feature importance score of the random forest, the top-ranked features are selected. The best feature subset is selected based on the feature importance score and input into the Logistic regression model for training to obtain the trained fault warning model.

[0093] In summary, the wind turbine generator data processing system and method provided in this application demonstrate higher accuracy in fault early warning of wind turbine generator drivetrain. It can effectively extract key features of the original input data, and has comprehensive fault early warning and control functions. It can reliably obtain fault early warning results, has high working efficiency, higher fitting degree, smaller error, and can more accurately predict faults in wind turbine generator drivetrain, thus achieving early warning.

[0094] Secondly, embodiments of this application provide a data processing method for wind turbine generator sets. (See reference...) Figure 2 This includes the following processes.

[0095] Step S1: Collect and upload multi-dimensional operating data of the wind turbine generator set in real time.

[0096] Step S2 involves preprocessing and normalizing the multidimensional operational data, and extracting key features from the processed data using a convolutional neural network to obtain key feature data. Specifically, the multidimensional operational data is decompressed, and missing value imputation is performed on the decompressed multidimensional operational data according to the physical characteristics of different measurement points to obtain complete multidimensional operational data. The quartile method is used to calculate the quartiles in the complete multidimensional operational data, and the upper and lower limits of outliers in the data are determined based on the quartiles. Complete multidimensional operational data outside the range of the quartiles are then removed to obtain normal multidimensional operational data. The normal multidimensional operational data is then normalized to obtain data to be characterized.

[0097] The quartile method involves: calculating the first and third quartiles of the multidimensional running data; calculating the interquartile range based on the first and third quartiles; determining the upper and lower limits of outliers based on the interquartile range; identifying and marking outliers from the multidimensional running data based on the upper and lower limits; and processing the marked outliers by deleting, replacing, or formatting them.

[0098] Step S3: Obtain the historical dataset and the Logistic regression model, and train the Logistic regression model based on the historical dataset to obtain the fault warning model. The historical dataset includes multi-dimensional operational data collected in the past.

[0099] Specifically, a pre-constructed convolutional neural network architecture is obtained, which includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer; the input layer receives data to be characterized, the convolutional layer extracts specific features from the data to be characterized to obtain a key feature matrix; the pooling layer performs dimensionality reduction on the key feature matrix; and the fully connected layer maps the dimensionality-reduced key feature matrix to the final output category to obtain the key feature data.

[0100] In a preferred embodiment, the method further includes: selecting an appropriate SVM kernel function according to user needs, and adjusting the parameters of the SVM using grid search and cross-validation methods to obtain an initial SVM model; obtaining a labeled training dataset to train the SVM model, and using a test dataset to evaluate the performance of the SVM model; using the trained SVM model to predict the fault category of the key feature data, and using a nonlinear kernel function for optimal calculation to obtain fault attribute classifications for different key feature data.

[0101] Step S4: Calculate the key feature data using the fault early warning model to obtain the probability of fault occurrence. When the probability of fault occurrence is greater than a preset threshold, issue an early warning signal.

[0102] The construction of the fault warning model includes: obtaining the risk factors for the occurrence of faults and the corresponding risk factor coefficients; constructing an initial fault warning model based on the risk factors, risk factor coefficients, and a Logistic regression model; solving the coefficients of the initial fault warning model using the maximum likelihood estimation method to obtain regression coefficients; and substituting the regression coefficients into the initial fault warning model to obtain the fault warning model.

[0103] In a more specific embodiment, the wind turbine generator data processing method adopts the following technical solution:

[0104] S01, The data acquisition module is responsible for collecting the operating data of the wind turbine generator set using sensors, including but not limited to multi-dimensional data collected by wind speed sensors, temperature sensors and vibration sensors.

[0105] S02. The preprocessing unit performs preliminary cleaning of the data acquired by the data acquisition module, removes or marks missing values ​​and obviously erroneous data points, and reasonably interpolates or removes null values ​​in the data according to the physical characteristics of different measurement points.

[0106] S03. The preprocessing unit uses the quartile method to process outlier data, marking and processing all data points below or above the upper and lower limits as outliers. Through the SCADA data parameter selection method, information redundancy is eliminated to avoid the model training getting stuck in local minima or gradient explosion. The cleaned data is normalized to eliminate the influence of different dimensions between wind turbine parameters, accelerate the gradient descent speed during model training, improve prediction accuracy, and make it suitable for subsequent data processing and analysis.

[0107] S04. The feature extraction unit inputs the data normalized by the preprocessing unit into the convolutional neural network. It obtains multiple convolutional feature maps through the convolution operation of the convolution kernel to improve the feature extraction capability. Through feature compression of the pooling layer, it reduces the model training parameters and obtains the main features.

[0108] S05. The feature extraction unit uses the support vector machine algorithm and nonlinear kernel function to classify the extracted key features in order to identify the data attributes related to fault warning, deeply explore the multi-dimensional fault warning and control patterns in the data, and realize the classification of fault warning data attributes.

[0109] S06. Fault early warning model: Construct a Logistic regression model and use the maximum likelihood estimation method to solve for the regression coefficients;

[0110] S07. Use cross-validation to evaluate the generalization ability of the fault warning model;

[0111] S08. Input the extracted key features into the trained Logistic regression model to calculate the probability of the fault occurring;

[0112] S09. Based on the calculated probability of failure, complete the fault identification. If the probability of failure exceeds the predetermined threshold, issue an early warning.

[0113] The data iteration module uses the data classified by the data processing module to build a historical dataset, and successively replaces the initial data. Based on the feature importance score of the random forest, it selects the top-ranked features, selects the best feature subset based on the feature importance score, and inputs it into the Logistic regression model for training to obtain the trained fault warning model. The specific steps include the following:

[0114] Create a historical dataset:

[0115] Data collection: Data collected for the first time or in previous years constitutes a historical dataset;

[0116] Data preparation: Historical data is preprocessed, including data cleaning, missing value handling, feature engineering, etc., to ensure data quality;

[0117] Feature extraction: Extracting useful features from historical data, such as vibration frequency, temperature changes, and current fluctuations;

[0118] Model training: Train the initial fault warning model using historical datasets;

[0119] Real-time data processing and early warning:

[0120] Data acquisition: Real-time acquisition of wind turbine generator operating data via sensors;

[0121] Data preprocessing: The collected real-time data is preprocessed, including data cleaning and normalization, to ensure that the real-time data is consistent with the format of the historical dataset.

[0122] Input Model: Input the preprocessed real-time data into the pre-trained Logistic Regression model;

[0123] Calculate the probability: Use the model to predict real-time data and calculate the warning probability for each sample;

[0124] Warning output: Warning output is generated based on the calculated probability. If the probability exceeds the preset threshold, a warning signal is issued.

[0125] Data filtering and analysis:

[0126] Data filtering: Based on the warning probability output by the model, a threshold (e.g., 0.8) is set, and data above this threshold are used as candidate warning data. The candidate warning data is initially filtered to remove obviously erroneous or abnormal data points, check the correlation between the warning data and other known fault data, and exclude isolated events or atypical situations.

[0127] Case Analysis: For each early warning case, review historical data to see if similar cases have occurred, and examine the situation and handling measures at that time. Analyze the early warning data from multiple dimensions, such as time series analysis, trend analysis, and pattern matching, to see the trend of the early warning data over time, whether there are periodic or sudden changes, analyze the trend of the early warning data over time, and determine whether it shows an upward or downward trend. Compare the early warning data with known fault modes to find similarities.

[0128] Validity assessment: Deep neural network models are used to extract features and classify early warning data to improve the accuracy of assessment;

[0129] Multi-source information fusion: The early warning data is fused with multi-source information (such as environmental data, operation records, etc.) to comprehensively judge the effectiveness of the early warning. Association rule mining technology is used to analyze the correlation between early warning data and other relevant factors, and the contextual information of the early warning is considered, such as weather conditions and equipment operating status, to improve the comprehensiveness of the judgment.

[0130] Results Confirmation: Summarize all analysis results, determine which warnings are true and valid, record the warning data that are confirmed to be true and valid to prepare for subsequent data updates and model iterations, and establish a feedback mechanism to record the warning data that are confirmed to be invalid in order to improve the model in the future.

[0131] Data updates and model iterations:

[0132] Data cleaning: Perform necessary cleaning on new data to remove noise and outliers;

[0133] Data balancing: Check if there is a class imbalance problem in the newly merged dataset, count the number of data in each class, and determine if there is a serious imbalance. If data imbalance is detected, take appropriate data balancing strategies. For minority class data, oversample to increase its quantity, simply copy minority class data to increase its quantity, or generate synthetic samples through interpolation to increase minority class data. For majority class data, undersample to reduce its quantity, randomly delete a portion of majority class data, and use the nearest neighbor algorithm to delete samples similar to majority class data. Combine oversampling and undersampling methods to achieve data balance. For minority class data, oversample and undersample majority class data. First, use SMOTE to generate synthetic samples, and then use nearest neighbor undersampling to reduce majority class data.

[0134] Model retraining:

[0135] Model loading: Load the existing model framework and prepare it for training;

[0136] Feature selection: The importance of existing features is evaluated using the feature importance score in random forest, correlation analysis is performed, highly correlated features are removed to avoid multicollinearity, and recursive feature elimination method is used to gradually remove unimportant features until the best feature combination is found. Based on the characteristics of the new dataset, new features are introduced or existing features are transformed to capture more effective information.

[0137] Hyperparameter tuning: Define the candidate range of hyperparameters, try all possible combinations one by one using grid search to find the optimal hyperparameter configuration, define a reasonable candidate value range for each hyperparameter, use cross-validation within the candidate range to evaluate the effect of each hyperparameter group, select the best hyperparameter combination based on the cross-validation results, randomly select hyperparameter combinations from the candidate range for evaluation, conduct multiple experiments on randomly selected hyperparameter combinations from the candidate range to reduce the influence of randomness, select the best hyperparameter combination based on the experimental results, use Bayesian optimization methods, construct a probabilistic model to estimate the optimal value of the hyperparameters, construct a Gaussian process or other probabilistic model to estimate the effect of the hyperparameters, define an objective function that maximizes the model's performance index, optimize the hyperparameters iteratively, select the hyperparameter combination most likely to improve performance for evaluation each time, set convergence conditions, and stop optimization when a threshold is reached or after a certain number of iterations.

[0138] The generalization ability of the model is evaluated using cross-validation, and the specific steps are as follows:

[0139] Split the dataset:

[0140] The historical dataset was divided into a training set and a test set, with 80% of the data used for training and 20% used for testing.

[0141] When performing cross-validation, the training set is further divided into K subsets (K = 5 or 10);

[0142] Perform K-fold cross-validation:

[0143] Select K-1 subsets from the K subsets as training data, and use the remaining subset as validation data;

[0144] Train a Logistic regression model using K-1 subsets of data;

[0145] Evaluate the model's performance using the remaining subset (validation set);

[0146] Repeat the above steps K times, each time selecting a different subset as the validation set;

[0147] Collect performance metrics for each validation (such as accuracy, precision, recall, F1 score, etc.);

[0148] Calculate average performance:

[0149] Calculate the average performance metric across K validations to evaluate the model's overall generalization ability.

[0150] It can calculate comprehensive metrics such as average accuracy, average precision, and average recall.

[0151] Choose the best model:

[0152] Based on the results of cross-validation, select the set of regression coefficients with the best performance;

[0153] This set of regression coefficients is the optimal solution obtained through a comprehensive analysis of all K validation results;

[0154] Final model evaluation:

[0155] The final selected model is evaluated using a test set (data that was not previously used in training).

[0156] This step is to further validate the model's performance on completely independent data.

[0157] Through the above steps, the present invention can achieve early warning of transmission chain failures in wind turbine generator sets, thereby effectively reducing the risk of transmission chain operation and improving the operating efficiency and safety of wind turbine generator sets.

[0158] In summary, this invention provides a complete solution for early warning of transmission chain faults in wind turbine generators by integrating data acquisition, preprocessing, feature extraction, fault early warning modeling, and early warning output, combined with edge computing technology and artificial intelligence algorithms.

[0159] Experimental preparation:

[0160] To verify the effectiveness of the wind turbine generator drivetrain fault early warning algorithm based on edge AI computing architecture proposed in this invention, the following experiment was conducted. A wind turbine generator fault simulation experimental platform was built, consisting of a frequency converter, a fault simulation test bench, a data acquisition instrument, and acquisition software. The high-speed end of the frequency converter was set to 1800 r / min with a sampling frequency of 512 kHz. The motor speed was set to 1000 r / min with a sampling frequency of 10.24 kHz. Data was acquired under different drivetrain conditions, with 250 sets of data collected for each condition, totaling 1000 sets. The fault early warning targets were wind turbine generators from major domestic manufacturers. Through continuous operation and on-site testing over the past year, 96 actual on-site fault records have been detected to date.

[0161] Results analysis:

[0162] The wind turbine generator drive chain fault early warning algorithm based on edge AI computing architecture proposed in this invention is compared with the fault early warning method of scraper conveyor chain drive system based on LSTM-Adam and the fault detection and diagnosis method of scraper conveyor chain drive system based on ANSYS. The actual fault early warning results are compared using the method of this invention, the LSTM-Adam method and the ANSYS method respectively. The application status of wind turbine generator drive chain health early warning is shown in Table 1.

[0163] Table 1: Application Status of Wind Turbine Generator Drivetrain Health Early Warning System

[0164]

[0165]

[0166] As shown in Table 1, the wind turbine generator drivetrain fault early warning algorithm based on edge AI computing architecture proposed in this invention can predict faults in the gearbox and main bearing. In contrast, the LSTM-Adam method can only predict abnormal data from the main bearing temperature sensor and the non-drive end temperature sensor of the gearbox intermediate shaft, while the ANSYS method can only predict abnormal temperatures in the gearbox oil sump, the main bearing temperature sensor, and the non-drive end temperature sensor of the gearbox intermediate shaft. Therefore, the wind turbine generator drivetrain fault early warning algorithm based on edge AI computing architecture designed in this invention can be considered feasible.

[0167] Table 2 Comparison of the accuracy of proactive early warning methods

[0168]

[0169] The proactive early warning accuracy of the method of this invention is higher than 80%. In the actual operation and maintenance application of wind turbine generator drivetrain, it can effectively extract the key features of the original input data. In contrast, the proactive early warning accuracy of the other two methods is lower. This proves that the fault early warning curve of the method of this invention has a higher fitting degree with the actual curve, smaller error, and superior performance. It has a better fault early warning effect and can realize early warning of wind turbine generator drivetrain. Therefore, the method of this invention has a comprehensive fault early warning control function, can reliably obtain fault early warning results, and has extremely high working efficiency in application, meeting the application needs of users for fault early warning control of wind turbine generator drivetrain.

[0170] Thirdly, embodiments of this application provide an electronic device, Figure 3 This is a block diagram illustrating an electronic device according to an exemplary embodiment. (e.g.) Figure 3 As shown, the electronic device may include a processor 11 and a memory 12 storing computer program instructions.

[0171] Specifically, the processor 11 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0172] The memory 12 may include a large-capacity storage device for data or instructions. For example, and not limitingly, the memory 12 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 12 may include removable or non-removable (or fixed) media. Where appropriate, the memory 12 may be internal or external to a data processing device. In a particular embodiment, the memory 12 is non-volatile memory. In a particular embodiment, the memory 12 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0173] The memory 12 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 11.

[0174] The processor 11 reads and executes computer program instructions stored in the memory 12 to implement any of the wind turbine generator data processing methods in the above embodiments.

[0175] In one embodiment, the electronic device may further include a communication interface 13 and a bus 10. Wherein, as... Figure 3 As shown, the processor 11, memory 12, and communication interface 13 are connected through bus 10 and communicate with each other.

[0176] The communication interface 13 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of this application. The communication port 13 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0177] Bus 10 includes hardware, software, or both, that couples components of an electronic device together. Bus 10 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 10 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 10 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0178] Fourthly, embodiments of this application provide a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the wind turbine generator data processing method provided in the first aspect.

[0179] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0180] In a possible implementation, the present invention can also be implemented as a program product comprising program code, which, when the program product is run on a terminal device, is used to cause the terminal device to perform steps implementing the wind turbine generator data processing method provided in the first aspect.

[0181] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0182] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0183] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for processing data from a wind turbine generator set, characterized in that, include: Real-time acquisition and uploading of multi-dimensional operational data from wind turbine generators; The multidimensional operational data undergoes preprocessing and normalization, including: The multidimensional operational data is decompressed, and missing values ​​are imputed in the decompressed multidimensional operational data according to the physical characteristics of different measurement points to obtain complete multidimensional operational data. The process involves calculating the quartiles of the complete multidimensional operational data using the quartile method, determining the upper and lower limits of outliers in the data based on the quartiles, and removing complete multidimensional operational data outside these limits to obtain normal multidimensional operational data. This includes: calculating the first and third quartiles of the multidimensional operational data; calculating the interquartile range based on the first and third quartiles; determining the upper and lower limits of outliers based on the interquartile range; identifying and marking outliers from the multidimensional operational data based on the upper and lower limits; and processing the marked outliers by deletion, replacement, or format transformation. The normalized multidimensional operating data is processed to obtain the data to be characterized; The key feature data is obtained by extracting key features from the processed data using a convolutional neural network, including: acquiring a pre-constructed convolutional neural network architecture, which includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer; receiving the data to be characterized through the input layer; extracting specific features from the data to be characterized through the convolutional layer to obtain a key feature matrix; performing dimensionality reduction on the key feature matrix through the pooling layer; and mapping the dimensionality-reduced key feature matrix to the final output category through the fully connected layer to obtain the key feature data. Obtain a historical dataset and a Logistic regression model, train the Logistic regression model based on the historical dataset to obtain a fault early warning model, wherein the historical dataset includes multi-dimensional operational data collected in the past; The fault early warning model calculates the probability of fault occurrence by analyzing the key feature data. When the probability of fault occurrence is greater than a preset threshold, an early warning signal is issued. The construction of the fault early warning model includes: obtaining the risk factors for the occurrence of faults and the corresponding risk factor coefficients; constructing an initial fault early warning model based on the risk factors, risk factor coefficients and a Logistic regression model; solving the coefficients of the initial fault early warning model using the maximum likelihood estimation method to obtain regression coefficients; and substituting the regression coefficients into the initial fault early warning model to obtain the fault early warning model.

2. The wind turbine generator data processing method according to claim 1, characterized in that, The method further includes: Select the appropriate SVM kernel function according to user needs, and use grid search and cross-validation methods to adjust the parameters of the SVM to obtain the initial SVM model; Obtain a labeled training dataset to train the SVM model, and use a test dataset to evaluate the performance of the SVM model. The trained SVM model is used to predict the fault category of the key feature data, and a nonlinear kernel function is used for optimal calculation to obtain the fault attribute classification of different key feature data.

3. A data processing system for wind turbine generator sets, characterized in that, The system is used to operate the wind turbine generator data processing method according to any one of claims 1-2, the system comprising: The data acquisition module is used to collect and upload multi-dimensional operating data of the wind turbine generator in real time, and send the multi-dimensional operating data to the data processing module. The data processing module is used to preprocess and normalize the multidimensional operating data, and extract key features from the preprocessed data through a convolutional neural network to obtain key feature data and send it to the fault warning module. The data iteration module is used to acquire historical datasets and a Logistic regression model, train the Logistic regression model based on the historical datasets, obtain the fault warning model, and send it to the fault warning module. The historical datasets include multi-dimensional operational data collected in the past. The fault early warning module is used to receive the key feature data and run the fault early warning model. The fault early warning model is used to calculate the key feature data to obtain the probability of fault occurrence. When the probability of fault occurrence is greater than a preset threshold, an early warning signal is issued.

4. The wind turbine generator data processing system according to claim 3, characterized in that, The data acquisition module includes a cache unit, which is communicatively connected to multiple acquisition points of the wind turbine generator set. The real-time acquisition and uploading of multi-dimensional operating data of the wind turbine generator set, and the sending of the multi-dimensional operating data to the data processing module, includes: The cache unit is monitored in real time. When new collected data is detected to be recorded, the new multidimensional running data is compressed and encrypted to obtain multidimensional running data. The data acquisition module uses a low-latency transmission protocol and multiplexing to transmit the multidimensional operational data from the acquisition point to the data processing module. An anomaly detection mechanism is used to monitor the data transmission process. If a data transmission failure is detected, the multidimensional operational data is resent.

5. The wind turbine generator data processing system according to claim 3, characterized in that, The data processing module includes a preprocessing unit and a feature extraction unit, wherein, The preprocessing unit is used to preprocess and normalize the multidimensional operational data; The feature extraction unit is used to extract key features from the preprocessed data using a convolutional neural network.

6. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the wind turbine generator data processing method as described in any one of claims 1 to 2.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the wind turbine generator data processing method as described in any one of claims 1 to 2.

Images