Artificial Intelligence-Based Wind Turbine Operation Status Monitoring Method

Abstract

This invention discloses an artificial intelligence-based method for monitoring the operating status of wind turbines. The method includes data acquisition, data preprocessing, construction of a stacked sample convolutional interactive network model, construction of an operating status monitoring model, design of anomaly threshold intervals, and monitoring of wind turbine operating status. This invention relates to the field of wind turbine operating status monitoring technology, specifically to an artificial intelligence-based method for monitoring wind turbine operating status. This invention obtains raw wind turbine operating data through data acquisition; employs data preprocessing methods such as data cleaning, data denoising, feature selection, data normalization, and dataset segmentation; uses a stacked sample convolutional interactive network model for operating status monitoring, enhancing the model's prediction accuracy; uses a federated learning framework to train the operating status monitoring model, effectively ensuring data privacy; and employs a Monte Carlo dropout and kernel density estimation-based method for calculating anomaly threshold intervals, enabling dynamic adjustment of the anomaly threshold intervals.

Description

Technical Field

[0001] This invention relates to the field of wind turbine operation status monitoring technology, specifically to a wind turbine operation status monitoring method based on artificial intelligence. Background Technology

[0002] Wind turbine operation status monitoring refers to the process of real-time monitoring and analysis of various key parameters and equipment status of wind turbines during operation. Through sensors and monitoring systems, data such as vibration, temperature and speed of wind turbines can be continuously tracked, and abnormalities and potential faults can be detected using the data. This not only improves the operating efficiency of the equipment but also extends its service life, thereby maximizing the economic benefits and reliability of the wind power generation system.

[0003] However, traditional wind turbine operation status monitoring methods suffer from several technical problems. They focus only on feature relationships in low-dimensional spatial domains and extract features based on a single temporal resolution, making it difficult to effectively utilize features at different scales. Furthermore, traditional wind turbine operation status monitoring methods face challenges such as the inability to collect sufficient and high-quality training data for a single wind turbine in a short period of time, and data privacy issues leading to low model accuracy. Additionally, traditional wind turbine operation status monitoring methods use static thresholds for abnormal status, inevitably introducing subjective uncertainties, resulting in poor monitoring performance and a high false alarm rate. Summary of the Invention

[0004] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an artificial intelligence-based wind turbine operation status monitoring method. Addressing the limitations of traditional wind turbine operation status monitoring methods, which focus only on feature relationships in low-dimensional spatial domains and extract features based on a single temporal resolution, making it difficult to effectively utilize features at different scales, this solution creatively employs a stacked sample convolutional interactive network model for operation status monitoring. This model can fully extract and fuse multi-dimensional and multi-resolution spatiotemporal features, enhancing the model's prediction accuracy. Furthermore, traditional wind turbine operation status monitoring methods suffer from limitations such as the inability to collect sufficient and high-quality training data from a single wind turbine within a short timeframe, and data privacy concerns, leading to limitations in model performance. To address the technical issue of low accuracy in model training, this solution creatively employs a federated learning framework to train the operational status monitoring model. This effectively ensures data privacy while fully utilizing data collected from different wind turbines, enabling faster acquisition of sufficient and high-quality training data. Furthermore, to address the technical problem of traditional wind turbine operational status monitoring methods using static thresholds for anomalies, which inevitably introduces subjective uncertainty and leads to poor monitoring performance and a high false alarm rate, this solution creatively adopts a method for calculating the anomaly threshold range based on Monte Carlo dropout and kernel density estimation. This effectively quantifies the uncertainty in the prediction process, dynamically adjusts the anomaly threshold range, and improves the model's accuracy.

[0005] The technical solution adopted by this invention is as follows: The wind turbine operation status monitoring method based on artificial intelligence provided by this invention includes the following steps:

[0006] Step S1: Data Acquisition;

[0007] Step S2: Data preprocessing;

[0008] Step S3: Constructing a stacked sample convolutional interactive network model;

[0009] Step S4: Construction of the operational status monitoring model;

[0010] Step S5: Design of abnormal state threshold range;

[0011] Step S6: Monitor the operating status of the wind turbine.

[0012] Furthermore, in step S1, the data acquisition is used to collect the raw data required for wind turbine operation status monitoring. Specifically, it involves obtaining multiple raw data sets of wind turbine operation data from the monitoring, control and data acquisition system of multiple wind turbines in the same wind farm.

[0013] The multiple wind turbine operation raw datasets are specifically defined as one wind turbine operation raw dataset for each wind turbine generator. The wind turbine operation raw datasets specifically include operating status data, environmental data, and operating cycle data. The operating status data specifically includes speed, torque, power generation, lubricating oil level, and vibration data. The environmental data specifically includes wind speed, temperature, air pressure, and humidity. The operating cycle data specifically includes cumulative operating time and number of start-stop cycles.

[0014] Further, in step S2, the data preprocessing is used to preprocess the collected raw data, specifically including the following steps:

[0015] Step S21: Data cleaning, used to clean the raw data, specifically to process the missing and duplicate values ​​in each of the original datasets of the wind turbine operation, to obtain multiple coarsely processed datasets;

[0016] Step S22: Data denoising, used to denoise the coarsely processed data, specifically by using the Kalman filter algorithm to denoise each of the coarsely processed datasets to obtain multiple denoised datasets;

[0017] Step S23: Feature selection, used to select features, specifically by using the method of calculating the Pearson correlation coefficient to select features for each of the denoised datasets, to obtain multiple wind turbine operation feature sets;

[0018] Step S24: Data normalization, used to normalize the wind turbine operating characteristics, specifically by using the min-max method to normalize the data of each wind turbine operating characteristic set, resulting in multiple normalized datasets;

[0019] Step S25: Dataset splitting, used to split the dataset, specifically to split each of the normalized datasets to obtain multiple wind turbine status training sets and multiple wind turbine status test sets;

[0020] Step S26: Preprocessing, specifically, by performing data cleaning, data denoising, feature selection, data normalization, and dataset segmentation on the original datasets of the multiple wind turbine operations to obtain multiple wind turbine status training sets and multiple wind turbine status test sets.

[0021] Further, in step S3, the construction of the stacked sample convolutional interaction network model, specifically includes the following steps:

[0022] Step S31: Design a coordinate attention mechanism, including the following steps:

[0023] Step S311: Coordinate attention pooling, specifically, using a one-dimensional average pooling operation with a kernel of (1,W) and a one-dimensional average pooling operation with a kernel of (H,1) on the input feature map of the coordinate attention mechanism. Processing is carried out, among which, Let C represent the input feature map of the coordinate attention mechanism, H represent the height of the input feature map of the coordinate attention mechanism, and W represent the width of the input feature map of the coordinate attention mechanism. The pooling operation output for each channel is calculated using the following formula:

[0024] ;

[0025] In the formula, This represents the output of the pooling operation at height h on the c-th channel. Let H represent the output of the pooling operation with width w on the c-th channel, where H represents the height of the input feature map for the coordinate attention mechanism, and W represents the width of the input feature map for the coordinate attention mechanism. This represents the value at coordinate (h, i) of the input feature map in the c-th channel of the coordinate attention mechanism. This represents the value of the input feature map at coordinate (j, w) in the c-th channel of the coordinate attention mechanism;

[0026] Step S312: Coordinate attention feature fusion, using the following formula:

[0027] ;

[0028] In the formula, This represents the coordinate attention fusion feature of the c-th channel. This represents the sigmoid function. This represents a convolution operation function with a kernel size of 1×1. Indicates a splicing operation;

[0029] Step S313: Calculate the weight matrix. Specifically, the coordinate attention fusion feature is divided into two independent tensors along the spatial dimension, and the weight matrix is ​​calculated using the following formula:

[0030] ;

[0031] In the formula, Represents the coordinate attention weight matrix. The tensor representing the coordinate attention fusion feature partition along the height is... The tensor representing the coordinate attention fusion feature partitioned along its width;

[0032] Step S314: Calculate the coordinates, noting the output mechanism. The formula used is as follows:

[0033] ;

[0034] In the formula, This indicates that the coordinates are output with attention to the mechanism. This represents element-wise multiplication;

[0035] Step S32: Construct the sample convolutional interaction submodule, including the following steps:

[0036] Step S321: Perform parity extraction, specifically by using parity extraction to downsample the input time series of the submodule into an odd subsequence and an even subsequence;

[0037] Step S322: Interactive learning, using the following formula:

[0038] ;

[0039] In the formula, Represents odd subsequences, Represents an even sequence. Represents the hyperbolic tangent function. Indicates the kernel size as One-dimensional convolution operation function, Indicates the kernel size as One-dimensional convolution operation function, This represents the Mish activation function. This represents the odd subsequence after interactive learning. This represents the even sequence after interactive learning;

[0040] Step S323: Calculate the submodule output using the following formula:

[0041] ;

[0042] In the formula, This indicates the output of the odd subsequence of the submodule. This indicates the output of the even sequence of the submodule. Indicates the kernel size as One-dimensional convolution operation function, Indicates the kernel size as One-dimensional convolution operation function;

[0043] Step S33: Construct a coordinate attention sample convolution interaction module. Specifically, the coordinate attention sample convolution interaction module is constructed by designing the coordinate attention mechanism and constructing the sample convolution interaction submodule.

[0044] Step S34: Construct a stacked sample convolutional interaction network model. Specifically, this involves constructing an encoder structure and a decoder structure to build the stacked sample convolutional interaction network model, thereby obtaining a running status monitoring model. The encoder structure specifically includes a multi-layer coordinate attention sample convolutional interaction module and a connection module. The decoder structure is specifically a fully connected layer with residual connections. The multi-layer coordinate attention sample convolutional interaction module is specifically arranged according to a binary tree structure. The connection module is used to connect the output of the multi-layer coordinate attention sample convolutional interaction module.

[0045] Step S35: Calculate the model loss using the following formula:

[0046] ;

[0047] In the formula, Loss represents the model loss, T represents the length of the time series, and L represents the total number of layers in the multi-coordinate convolution interaction module. This represents the t-th predicted value of layer l. This represents the t-th actual value.

[0048] Furthermore, in step S4, the operation status monitoring model construction is used to construct the model required for wind turbine operation status monitoring, specifically by constructing a federated learning framework to obtain the operation status monitoring model;

[0049] The construction of the operational status monitoring model specifically includes the following steps:

[0050] Step S41: Construct the main model and sub-models, specifically by constructing a single main model and multiple sub-models. The main model adopts the stacked sample convolutional interaction network model, and the sub-models adopt the stacked sample convolutional interaction network model.

[0051] Step S42: Sub-model training, specifically, each sub-model is trained using the wind turbine state training set corresponding to a wind turbine, the wind turbine state test set is used to evaluate the model performance, and the trained sub-model parameters are sent to the main model;

[0052] Step S43: Update the main model parameters, specifically by updating the main model parameters based on the sub-model parameters, and sending the updated main model parameters to the sub-model for a new round of training. The formula used to update the main model parameters is as follows:

[0053] ;

[0054] In the formula, This represents the main model parameters in the (d+1)th training epoch. This represents the sub-model parameters of the nth sub-model in the d-th training epoch. M represents the amount of data in the wind turbine state training set of the nth sub-model, M represents the total amount of data in all wind turbine state training sets, N represents the total number of sub-models, and d represents the current training round.

[0055] Step S44: Construct a runtime status monitoring model. Specifically, this involves constructing a main model and sub-models, training the sub-models, and updating the main model parameters to build a federated learning framework and train it. The trained main model is then used as the runtime status monitoring model.

[0056] Further, in step S5, the design of the state anomaly threshold interval, used to design the state anomaly threshold interval, specifically includes the following steps:

[0057] Step S51: Monte Carlo dropout design, specifically, performing multiple forward propagations for each input. In each forward propagation, the weights of the running status monitoring model are treated as random variables and follow a Bernoulli distribution. The Monte Carlo dropout design uses the following formula.

[0058] ;

[0059] In the formula, Let represent the weights of the runtime status monitoring model at the q-th forward propagation, and p represent the probability that the runtime status monitoring model weights are retained. The weights of the running state monitoring model at the q-th forward propagation are represented. The probability that the weights of the operational status monitoring model are retained, following a Bernoulli distribution, is p;

[0060] Step S52: Calculate the model predictions using the following formula:

[0061] ;

[0062] In the formula, x represents the input of the operation status monitoring model. This represents the predicted value of the operational status monitoring model based on the input x after incorporating Monte Carlo dropout, where Q represents the total number of forward propagations. This represents the weight of the operation status monitoring model at the q-th forward propagation. The predicted value of the operating status monitoring model for the input x of the operating status monitoring model;

[0063] Step S53: Calculate the probability density function, specifically by... Considering the predicted value with respect to time step, the probability density function is calculated based on the kernel density estimation method, using the following formula:

[0064] ;

[0065] In the formula, Let b represent the probability density function, and b represent the bandwidth coefficient of the Gaussian kernel function. Represents the Gaussian kernel function. This indicates that the weights of the operation status monitoring model are the same as those at the time of the first forward propagation. The predicted value of the operating status monitoring model for the input x. This represents the weight of the operation status monitoring model at the Q-th forward propagation. The predicted value of the operating status monitoring model for the input x of the operating status monitoring model;

[0066] Step S54: Calculate the abnormal state threshold range using the following formula:

[0067] ;

[0068] In the formula, V represents the threshold range for abnormal states. This indicates the lower limit of the abnormal status threshold. This indicates the upper limit of the abnormal status threshold. Represents the probability density function Integrating yields the cumulative distribution function. , Represents the cumulative distribution function inverse function, Indicates the confidence level.

[0069] Further, in step S6, the wind turbine operation status monitoring specifically involves using the operation status monitoring model to monitor the wind turbine operation status. If the model prediction value of the operation status monitoring model does not continuously exceed the state abnormality threshold range, the wind turbine is operating normally. If the model prediction value of the operation status monitoring model continuously exceeds the state abnormality threshold range, the wind turbine is operating abnormally.

[0070] The beneficial effects achieved by the present invention using the above solution are as follows:

[0071] (1) In view of the technical problem that traditional wind turbine operation status monitoring methods only focus on the feature relationships in the low-dimensional spatial domain and extract features based on a single time resolution, making it difficult to effectively utilize features at different scales, this scheme creatively adopts a stacked sample convolutional interactive network model for operation status monitoring, which can fully extract and integrate multi-dimensional and multi-resolution spatiotemporal features, thereby enhancing the prediction accuracy of the model.

[0072] (2) In view of the technical problems of traditional wind turbine operation status monitoring methods, such as the inability of a single wind turbine to collect sufficient and high-quality training data in a short period of time, and the existence of data privacy, which leads to low model accuracy, this solution creatively adopts a federated learning framework to train the operation status monitoring model. It can effectively guarantee data privacy while making full use of the data collected by different wind turbines, and can obtain sufficient and high-quality training data more quickly.

[0073] (3) In view of the technical problems of traditional wind turbine operation status monitoring methods, which set the state abnormality threshold as a static threshold and inevitably introduce subjective uncertainty, resulting in poor monitoring performance and high false alarm rate, this solution creatively adopts a state abnormality threshold interval calculation method based on Monte Carlo discarding and kernel density estimation, which effectively quantifies the uncertainty in the prediction process, can dynamically adjust the state abnormality threshold interval, and improves the accuracy of the model. Attached Figure Description

[0074] Figure 1 A flowchart illustrating the artificial intelligence-based wind turbine operation status monitoring method provided by this invention;

[0075] Figure 2 This is a flowchart illustrating the data preprocessing process in step S2.

[0076] Figure 3 A flowchart illustrating the process of constructing the stacked sample convolutional interactive network model in step S3;

[0077] Figure 4 A flowchart illustrating the process of building the operational status monitoring model for step S4;

[0078] Figure 5A flowchart illustrating the design of the abnormal state threshold range for step S5.

[0079] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0080] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0081] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0082] Example 1, see Figure 1 The technical solution adopted by this invention is as follows: The wind turbine operation status monitoring method based on artificial intelligence provided by this invention includes the following steps:

[0083] Step S1: Data Acquisition;

[0084] Step S2: Data preprocessing;

[0085] Step S3: Constructing a stacked sample convolutional interactive network model;

[0086] Step S4: Construction of the operational status monitoring model;

[0087] Step S5: Design of abnormal state threshold range;

[0088] Step S6: Monitor the operating status of the wind turbine.

[0089] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the data acquisition is used to collect the raw data required for wind turbine operation status monitoring. Specifically, it is to obtain multiple raw data sets of wind turbine operation from the monitoring, control and data acquisition system of multiple wind turbines in the same wind farm through data acquisition.

[0090] The multiple wind turbine operation raw datasets are specifically defined as one wind turbine operation raw dataset for each wind turbine generator. The wind turbine operation raw datasets specifically include operating status data, environmental data, and operating cycle data. The operating status data specifically includes speed, torque, power generation, lubricating oil level, and vibration data. The environmental data specifically includes wind speed, temperature, air pressure, and humidity. The operating cycle data specifically includes cumulative operating time and number of start-stop cycles.

[0091] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to preprocess the collected raw data, specifically including the following steps:

[0092] Step S21: Data cleaning, used to clean the raw data, specifically to process the missing and duplicate values ​​in each of the original datasets of the wind turbine operation, to obtain multiple coarsely processed datasets;

[0093] Step S22: Data denoising, used to denoise the coarsely processed data, specifically by using the Kalman filter algorithm to denoise each of the coarsely processed datasets to obtain multiple denoised datasets;

[0094] Step S23: Feature selection, used to select features, specifically by using the method of calculating the Pearson correlation coefficient to select features for each of the denoised datasets, to obtain multiple wind turbine operation feature sets;

[0095] Step S24: Data normalization, used to normalize the wind turbine operating characteristics, specifically by using the min-max method to normalize the data of each wind turbine operating characteristic set, resulting in multiple normalized datasets;

[0096] Step S25: Dataset splitting, used to split the dataset, specifically to split each of the normalized datasets to obtain multiple wind turbine status training sets and multiple wind turbine status test sets;

[0097] Step S26: Preprocessing, specifically, by performing data cleaning, data denoising, feature selection, data normalization, and dataset segmentation on the original datasets of the multiple wind turbine operations to obtain multiple wind turbine status training sets and multiple wind turbine status test sets.

[0098] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the construction of the stacked sample convolutional interaction network model specifically includes the following steps:

[0099] Step S31: Design a coordinate attention mechanism, including the following steps:

[0100] Step S311: Coordinate attention pooling, specifically, using a one-dimensional average pooling operation with a kernel of (1,W) and a one-dimensional average pooling operation with a kernel of (H,1) on the input feature map of the coordinate attention mechanism. Processing is carried out, among which, Let C represent the input feature map of the coordinate attention mechanism, H represent the height of the input feature map of the coordinate attention mechanism, and W represent the width of the input feature map of the coordinate attention mechanism. The pooling operation output for each channel is calculated using the following formula:

[0101] ;

[0102] In the formula, This represents the output of the pooling operation at height h on the c-th channel. Let H represent the output of the pooling operation with width w on the c-th channel, where H represents the height of the input feature map for the coordinate attention mechanism, and W represents the width of the input feature map for the coordinate attention mechanism. This represents the value at coordinate (h, i) of the input feature map in the c-th channel of the coordinate attention mechanism. This represents the value of the input feature map at coordinate (j, w) in the c-th channel of the coordinate attention mechanism;

[0103] Step S312: Coordinate attention feature fusion, using the following formula:

[0104] ;

[0105] In the formula, This represents the coordinate attention fusion feature of the c-th channel. This represents the sigmoid function. This represents a convolution operation function with a kernel size of 1×1. Indicates a splicing operation;

[0106] Step S313: Calculate the weight matrix. Specifically, the coordinate attention fusion feature is divided into two independent tensors along the spatial dimension, and the weight matrix is ​​calculated using the following formula:

[0107] ;

[0108] In the formula, Represents the coordinate attention weight matrix. The tensor representing the coordinate attention fusion feature partition along the height is... The tensor representing the coordinate attention fusion feature partitioned along its width;

[0109] Step S314: Calculate the coordinates, noting the output mechanism. The formula used is as follows:

[0110] ;

[0111] In the formula, This indicates that the coordinates are output with attention to the mechanism. This represents element-wise multiplication;

[0112] Step S32: Construct the sample convolutional interaction submodule, including the following steps:

[0113] Step S321: Perform parity extraction, specifically by using parity extraction to downsample the input time series of the submodule into an odd subsequence and an even subsequence;

[0114] Step S322: Interactive learning, using the following formula:

[0115] ;

[0116] In the formula, Represents odd subsequences, Represents an even sequence. Represents the hyperbolic tangent function. Indicates the kernel size as One-dimensional convolution operation function, Indicates the kernel size as One-dimensional convolution operation function, This represents the Mish activation function. This represents the odd subsequence after interactive learning. This represents the even sequence after interactive learning;

[0117] Step S323: Calculate the submodule output using the following formula:

[0118] ;

[0119] In the formula, This indicates the output of the odd subsequence of the submodule. This indicates the output of the even sequence of the submodule. Indicates the kernel size as One-dimensional convolution operation function, Indicates the kernel size as One-dimensional convolution operation function;

[0120] Step S33: Construct a coordinate attention sample convolution interaction module. Specifically, the coordinate attention sample convolution interaction module is constructed by designing the coordinate attention mechanism and constructing the sample convolution interaction submodule.

[0121] Step S34: Construct a stacked sample convolutional interaction network model. Specifically, this involves constructing an encoder structure and a decoder structure to build the stacked sample convolutional interaction network model, thereby obtaining a running status monitoring model. The encoder structure specifically includes a multi-layer coordinate attention sample convolutional interaction module and a connection module. The decoder structure is specifically a fully connected layer with residual connections. The multi-layer coordinate attention sample convolutional interaction module is specifically arranged according to a binary tree structure. The connection module is used to connect the output of the multi-layer coordinate attention sample convolutional interaction module.

[0122] Step S35: Calculate the model loss using the following formula:

[0123] ;

[0124] In the formula, Loss represents the model loss, T represents the length of the time series, and L represents the total number of layers in the multi-coordinate convolution interaction module. This represents the t-th predicted value of layer l. This represents the t-th actual value.

[0125] By performing the above operations, this solution addresses the technical problem that traditional wind turbine operation status monitoring methods only focus on feature relationships in low-dimensional spatial domains and extract features based on a single temporal resolution, making it difficult to effectively utilize features at different scales. This solution creatively adopts a stacked sample convolutional interactive network model for operation status monitoring, which can fully extract and fuse multi-dimensional and multi-resolution spatiotemporal features, thereby enhancing the model's prediction accuracy.

[0126] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S4, the operation status monitoring model is constructed to build the model required for wind turbine operation status monitoring. Specifically, the operation status monitoring model is obtained by constructing a federated learning framework.

[0127] The construction of the operational status monitoring model specifically includes the following steps:

[0128] Step S41: Construct the main model and sub-models, specifically by constructing a single main model and multiple sub-models. The main model adopts the stacked sample convolutional interaction network model, and the sub-models adopt the stacked sample convolutional interaction network model.

[0129] Step S42: Sub-model training, specifically, each sub-model is trained using the wind turbine state training set corresponding to a wind turbine, the wind turbine state test set is used to evaluate the model performance, and the trained sub-model parameters are sent to the main model;

[0130] Step S43: Update the main model parameters, specifically by updating the main model parameters based on the sub-model parameters, and sending the updated main model parameters to the sub-model for a new round of training. The formula used to update the main model parameters is as follows:

[0131] ;

[0132] In the formula, This represents the main model parameters in the (d+1)th training epoch. This represents the sub-model parameters of the nth sub-model in the d-th training epoch. M represents the amount of data in the wind turbine state training set of the nth sub-model, M represents the total amount of data in all wind turbine state training sets, N represents the total number of sub-models, and d represents the current training round.

[0133] Step S44: Construct a runtime status monitoring model. Specifically, this involves constructing a main model and sub-models, training the sub-models, and updating the main model parameters to build a federated learning framework and train it. The trained main model is then used as the runtime status monitoring model.

[0134] By performing the above operations, this solution addresses the technical problems of traditional wind turbine operation status monitoring methods, such as the inability of a single wind turbine to collect sufficient and high-quality training data in a short period of time, and the existence of data privacy issues leading to low model accuracy. This solution creatively adopts a federated learning framework to train the operation status monitoring model, which can effectively guarantee data privacy while making full use of data collected from different wind turbines, and can obtain sufficient and high-quality training data more quickly.

[0135] Example 6, see Figure 1 and Figure 5 This embodiment is based on the above embodiment. In step S5, the design of the state anomaly threshold interval is used to design the state anomaly threshold interval, and specifically includes the following steps:

[0136] Step S51: Monte Carlo dropout design, specifically, performing multiple forward propagations for each input. In each forward propagation, the weights of the running status monitoring model are treated as random variables and follow a Bernoulli distribution. The Monte Carlo dropout design uses the following formula.

[0137] ;

[0138] In the formula, Let represent the weights of the runtime status monitoring model at the q-th forward propagation, and p represent the probability that the runtime status monitoring model weights are retained. The weights of the running state monitoring model at the q-th forward propagation are represented. The probability that the weights of the operational status monitoring model are retained, following a Bernoulli distribution, is p;

[0139] Step S52: Calculate the model predictions using the following formula:

[0140] ;

[0141] In the formula, x represents the input of the operation status monitoring model. This represents the predicted value of the operational status monitoring model based on the input x after incorporating Monte Carlo dropout, where Q represents the total number of forward propagations. This represents the weight of the operation status monitoring model at the q-th forward propagation. The predicted value of the operating status monitoring model for the input x of the operating status monitoring model;

[0142] Step S53: Calculate the probability density function, specifically by... Considering the predicted value with respect to time step, the probability density function is calculated based on the kernel density estimation method, using the following formula:

[0143] ;

[0144] In the formula, Let b represent the probability density function, and b represent the bandwidth coefficient of the Gaussian kernel function. Represents the Gaussian kernel function. This indicates that the weights of the operation status monitoring model are the same as those at the time of the first forward propagation. The predicted value of the operating status monitoring model for the input x. This represents the weight of the operation status monitoring model at the Q-th forward propagation. The predicted value of the operating status monitoring model for the input x of the operating status monitoring model;

[0145] Step S54: Calculate the abnormal state threshold range using the following formula:

[0146] ;

[0147] In the formula, V represents the threshold range for abnormal states. This indicates the lower limit of the abnormal status threshold. This indicates the upper limit of the abnormal status threshold. Represents the probability density function Integrating yields the cumulative distribution function. , Represents the cumulative distribution function inverse function, Indicates the confidence level.

[0148] By performing the above operations, this solution addresses the technical problems of traditional wind turbine operation status monitoring methods, which use static thresholds for abnormal status and inevitably introduce subjective uncertainties, leading to poor monitoring performance and a high false alarm rate. Instead, it creatively employs a method for calculating the abnormal status threshold range based on Monte Carlo discarding and kernel density estimation. This method effectively quantifies the uncertainties in the prediction process, dynamically adjusts the abnormal status threshold range, and improves the accuracy of the model.

[0149] Example 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the wind turbine operation status monitoring specifically involves using the operation status monitoring model to monitor the wind turbine operation status. If the model prediction value of the operation status monitoring model does not continuously exceed the state abnormality threshold range, the wind turbine is operating normally. If the model prediction value of the operation status monitoring model continuously exceeds the state abnormality threshold range, the wind turbine is operating abnormally.

[0150] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0151] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0152] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A wind turbine operation status monitoring method based on artificial intelligence, characterized in that: The method includes the following steps: Step S1: Data Acquisition; Step S2: Data preprocessing, the collected raw wind turbine operation data is cleaned, denoised, feature selected, normalized and split into multiple wind turbine status training sets and multiple wind turbine status test sets. Step S3: Construction of stacked sample convolutional interaction network model, specifically by designing a coordinate attention mechanism and constructing a sample convolutional interaction sub-module to obtain the stacked sample convolutional interaction network model; Step S4: Construction of the runtime status monitoring model, specifically by constructing a federated learning framework to obtain the runtime status monitoring model; Step S5: Design of the state anomaly threshold interval, specifically by using Monte Carlo dropout design and calculating the probability density function based on kernel density estimation to obtain the state anomaly threshold interval; Step S6: Wind turbine operation status monitoring, specifically, monitoring the wind turbine operation status through the operation status monitoring model, and determining the wind turbine operation status in conjunction with the abnormal status threshold range; The construction of the stacked sample convolutional interactive network model specifically includes the following steps: Step S31: Design a coordinate attention mechanism, including the following steps: Step S311: Coordinate attention pooling, specifically, using a one-dimensional average pooling operation with a kernel of (1,W) and a one-dimensional average pooling operation with a kernel of (H,1) on the input feature map of the coordinate attention mechanism. Processing is carried out, among which, Let C represent the input feature map of the coordinate attention mechanism, H represent the height of the input feature map of the coordinate attention mechanism, and W represent the width of the input feature map of the coordinate attention mechanism. The pooling operation output for each channel is calculated using the following formula: ； In the formula, This represents the output of the pooling operation at height h on the c-th channel. Let H represent the output of the pooling operation with width w on the c-th channel, where H represents the height of the input feature map for the coordinate attention mechanism, and W represents the width of the input feature map for the coordinate attention mechanism. This represents the value at coordinate (h, i) of the input feature map in the c-th channel of the coordinate attention mechanism. This represents the value of the input feature map at coordinate (j, w) in the c-th channel of the coordinate attention mechanism; Step S312: Coordinate attention feature fusion, using the following formula: ； In the formula, This represents the coordinate attention fusion feature of the c-th channel. This represents the sigmoid function. This represents a convolution operation function with a kernel size of 1×1. Indicates a splicing operation; Step S313: Calculate the weight matrix. Specifically, the coordinate attention fusion feature is divided into two independent tensors along the spatial dimension, and the weight matrix is ​​calculated using the following formula: ； In the formula, Represents the coordinate attention weight matrix. The tensor representing the coordinate attention fusion feature partition along the height is... The tensor representing the coordinate attention fusion feature partitioned along its width; Step S314: Calculate the coordinates, noting the output mechanism. The formula used is as follows: ； In the formula, This indicates that the coordinates are output with attention to the mechanism. This represents element-wise multiplication; Step S32: Construct the sample convolutional interaction submodule, including the following steps: Step S321: Perform parity extraction, specifically by using parity extraction to downsample the input time series of the submodule into an odd subsequence and an even subsequence; Step S322: Interactive learning, using the following formula: ； In the formula, Represents odd subsequences, Represents an even sequence. Represents the hyperbolic tangent function. Indicates the kernel size as One-dimensional convolution operation function, Indicates the kernel size as One-dimensional convolution operation function, This represents the Mish activation function. This represents the odd subsequence after interactive learning. This represents the even sequence after interactive learning; Step S323: Calculate the submodule output using the following formula: ； In the formula, This indicates the output of the odd sub-sequence of the submodule. This indicates the even sequence output of the submodule. Indicates the kernel size as One-dimensional convolution operation function, Indicates the kernel size as One-dimensional convolution operation function; Step S33: Construct a coordinate attention sample convolution interaction module. Specifically, the coordinate attention sample convolution interaction module is constructed by designing the coordinate attention mechanism and constructing the sample convolution interaction submodule. Step S34: Construct a stacked sample convolutional interaction network model. Specifically, this involves constructing an encoder structure and a decoder structure to build the stacked sample convolutional interaction network model, thereby obtaining a running status monitoring model. The encoder structure specifically includes a multi-layer coordinate attention sample convolutional interaction module and a connection module. The decoder structure is specifically a fully connected layer with residual connections. The multi-layer coordinate attention sample convolutional interaction module is specifically arranged according to a binary tree structure. The connection module is used to connect the output of the multi-layer coordinate attention sample convolutional interaction module. Step S35: Calculate the model loss using the following formula: ； In the formula, Loss represents the model loss, T represents the length of the time series, and L represents the total number of layers in the multi-coordinate convolution interaction module. This represents the t-th predicted value of layer l. This represents the t-th actual value.

2. The method for monitoring the operating status of wind turbines based on artificial intelligence according to claim 1, characterized in that: In step S4, the operation status monitoring model is constructed to build the model required for wind turbine operation status monitoring. Specifically, the operation status monitoring model is obtained by constructing a federated learning framework. The construction of the operational status monitoring model specifically includes the following steps: Step S41: Construct the main model and sub-models, specifically by constructing a single main model and multiple sub-models. The main model adopts the stacked sample convolutional interaction network model, and the sub-models adopt the stacked sample convolutional interaction network model. Step S42: Sub-model training, specifically, each sub-model is trained using the wind turbine state training set corresponding to a wind turbine, the wind turbine state test set is used to evaluate the model performance, and the trained sub-model parameters are sent to the main model; Step S43: Update the main model parameters, specifically by updating the main model parameters based on the sub-model parameters, and sending the updated main model parameters to the sub-model for a new round of training. The formula used to update the main model parameters is as follows: ； In the formula, This represents the main model parameters in the (d+1)th training epoch. This represents the sub-model parameters of the nth sub-model in the d-th training epoch. M represents the amount of data in the wind turbine state training set of the nth sub-model, M represents the total amount of data in all wind turbine state training sets, N represents the total number of sub-models, and d represents the current training round. Step S44: Construct a runtime status monitoring model. Specifically, this involves constructing a main model and sub-models, training the sub-models, and updating the main model parameters to build a federated learning framework and train it. The trained main model is then used as the runtime status monitoring model.

3. The method for monitoring the operating status of wind turbines based on artificial intelligence according to claim 1, characterized in that: In step S5, the design of the state anomaly threshold interval, used to design the state anomaly threshold interval, specifically includes the following steps: Step S51: Monte Carlo dropout design, specifically, performing multiple forward propagations for each input. In each forward propagation, the weights of the running status monitoring model are treated as random variables and follow a Bernoulli distribution. The Monte Carlo dropout design uses the following formula. ； In the formula, Let represent the weights of the runtime status monitoring model at the q-th forward propagation, and p represent the probability that the runtime status monitoring model weights are retained. The weights of the running state monitoring model at the q-th forward propagation are represented. The probability that the weights of the operation status monitoring model are retained, following a Bernoulli distribution, is p; Step S52: Calculate the model predictions using the following formula: ； In the formula, x represents the input of the operation status monitoring model. This represents the predicted value of the operational status monitoring model based on the input x after incorporating Monte Carlo dropout, where Q represents the total number of forward propagations. This represents the weight of the operation status monitoring model at the q-th forward propagation. The predicted value of the operating status monitoring model for the input x of the operating status monitoring model; Step S53: Calculate the probability density function, specifically by... Considering the predicted value with respect to time step, the probability density function is calculated based on the kernel density estimation method, using the following formula: ； In the formula, Let b represent the probability density function, and b represent the bandwidth coefficient of the Gaussian kernel function. Represents the Gaussian kernel function. This indicates that the weights of the operation status monitoring model are the same as those at the time of the first forward propagation. The predicted value of the operating status monitoring model for the input x. This represents the weight of the operation status monitoring model at the Q-th forward propagation. The predicted value of the operating status monitoring model for the input x of the operating status monitoring model; Step S54: Calculate the abnormal state threshold range using the following formula: ； In the formula, V represents the threshold range for abnormal states. This indicates the lower limit of the abnormal status threshold. This indicates the upper limit of the abnormal status threshold. Represents the probability density function Integrating yields the cumulative distribution function. , Represents the cumulative distribution function inverse function, Indicates the confidence level.

4. The method for monitoring the operating status of wind turbines based on artificial intelligence according to claim 1, characterized in that: In step S1, the data acquisition is used to collect the raw data required for wind turbine operation status monitoring. Specifically, it involves obtaining multiple raw data sets of wind turbine operation data from the monitoring, control and data acquisition system of multiple wind turbines in the same wind farm. The multiple wind turbine operation raw datasets are specifically defined as one wind turbine operation raw dataset for each wind turbine generator. The wind turbine operation raw datasets specifically include operating status data, environmental data, and operating cycle data. The operating status data specifically includes speed, torque, power generation, lubricating oil level, and vibration data. The environmental data specifically includes wind speed, temperature, air pressure, and humidity. The operating cycle data specifically includes cumulative operating time and number of start-stop cycles.

5. The method for monitoring the operating status of wind turbines based on artificial intelligence according to claim 1, characterized in that: In step S2, the data preprocessing is used to preprocess the collected raw data, specifically including the following steps: Step S21: Data cleaning, used to clean the raw data, specifically to process the missing and duplicate values ​​in each of the original datasets of the wind turbine operation, to obtain multiple coarsely processed datasets; Step S22: Data denoising, used to denoise the coarsely processed data, specifically by using the Kalman filter algorithm to denoise each of the coarsely processed datasets to obtain multiple denoised datasets; Step S23: Feature selection, used to select features, specifically by using the method of calculating the Pearson correlation coefficient to select features for each of the denoised datasets, to obtain multiple wind turbine operation feature sets; Step S24: Data normalization, used to normalize the wind turbine operating characteristics, specifically by using the min-max method to normalize the data of each wind turbine operating characteristic set, resulting in multiple normalized datasets; Step S25: Dataset splitting, used to split the dataset, specifically to split each of the normalized datasets to obtain multiple wind turbine status training sets and multiple wind turbine status test sets; Step S26: Preprocessing, specifically, by performing data cleaning, data denoising, feature selection, data normalization, and dataset segmentation on the original datasets of the multiple wind turbine operations to obtain multiple wind turbine status training sets and multiple wind turbine status test sets.

6. The method for monitoring the operating status of wind turbines based on artificial intelligence according to claim 1, characterized in that: In step S6, the wind turbine operation status monitoring specifically involves using the operation status monitoring model to monitor the wind turbine operation status. If the model prediction value of the operation status monitoring model does not continuously exceed the state abnormality threshold range, the wind turbine is operating normally. If the model prediction value of the operation status monitoring model continuously exceeds the state abnormality threshold range, the wind turbine is operating abnormally.

Images