Method and system for predicting running state of wind turbine generator

Abstract

The invention relates to the technical field of wind generating sets, in particular to a method and system for predicting the running state of a wind generating set. According to the prediction method, operation data and environment data of the wind turbine generator are collected in multiple dimensions, key factors influencing the operation state of the generator are comprehensively covered, and prediction one-sidedness caused by a single data dimension is avoided; noise and abnormal values are eliminated through data preprocessing, and the data quality is effectively improved; key features are extracted through a sliding window method and a random forest algorithm, key information of data is reserved, and model training complexity is reduced; the constructed improved fusion prediction model combines the fitting advantage of the LSTM on time series data and the capturing capability of the XGBoost on a nonlinear relationship, the prediction precision is improved, and the accurate prediction of the operation state of the wind turbine generator is realized. The method effectively improves the accuracy and real-time performance of the operation state prediction of the wind turbine generator, provides a scientific basis for the operation and maintenance of the wind turbine generator, reduces the downtime of the wind turbine generator, reduces the operation and maintenance cost, and guarantees the stable operation of a wind power plant.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine technology, specifically relating to a method and system for predicting the operating status of wind turbines. Background Technology

[0002] As the global energy structure shifts towards cleaner and lower-carbon energy, wind power, as a crucial component of new energy generation, is experiencing continuous growth in both installed capacity and electricity generation. Wind turbines are typically installed in complex environments such as the field and offshore, and are subject to long-term operational challenges including wind speed variations, temperature fluctuations, and blade damage. Failure to accurately and promptly predict the operating status of wind turbines can lead to downtime for maintenance, resulting in significant economic losses and even safety accidents.

[0003] Existing methods for predicting the operating status of wind turbines mainly include physical model-based methods, data-driven methods, and rule-based methods. Physical model-based methods require establishing accurate physical models of each component of the wind turbine. However, wind turbines have complex structures and strong coupling relationships between components, making it difficult to construct comprehensive and accurate physical models, and they also have poor adaptability to environmental changes. Rule-based methods rely on the experience of maintenance personnel to formulate judgment rules, which are highly subjective, difficult to cover all operating scenarios, and have low prediction accuracy. While data-driven methods can utilize historical operating data for prediction, existing data-driven methods often use a single model, ignoring the temporal correlation of data and the synergistic effect of multi-dimensional features, easily leading to prediction lag and insufficient accuracy, making it difficult to meet the needs of efficient wind turbine operation and maintenance. Summary of the Invention

[0004] Based on the aforementioned problems in the existing technology, the purpose of this invention is to provide a method and system for predicting the operating status of wind turbine units.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for predicting the operating status of a wind turbine generator, comprising: Step S1: Collect the operating data and environmental data of the target wind turbine, and preprocess the operating data and environmental data to obtain a processed dataset; the operating data includes the speed data, power data, temperature data, vibration data, voltage and current data of the target wind turbine; the environmental data includes wind speed data, wind direction data, ambient temperature data, ambient humidity data, and atmospheric pressure data. Step S2: Extract features from the data in the processing dataset using the sliding window method, and filter key features using the random forest algorithm to obtain a key feature set; Step S3: Divide the key feature set into a training set, a validation set, and a test set according to the proportions; Step S4: Construct a fusion prediction model based on the fusion of LSTM network and XGBoost technology. Input the training set into the fusion prediction model for model training, and improve the training parameters of the fusion prediction model based on the training results to obtain the running state prediction model. Step S5: Input the test set into the operation status prediction model to obtain the operation status prediction result of the target wind turbine.

[0006] A further improvement of the present invention is that the step of preprocessing the running data and the environmental data in step S1 to obtain the processed dataset includes: The operating data and the environmental data are subjected to noise removal, outlier handling, and data normalization. The noise removal operation uses a wavelet transform algorithm to remove high-frequency noise from the running data and the environmental data. The outlier handling is based on the 3σ criterion to identify outliers in the operational data and the environmental data; The data normalization process uses Min-Max normalization to map the running data and the environmental data after the noise removal operation and the outlier processing to the [0,1] interval for data normalization.

[0007] A further improvement of the present invention is that the rotational speed data includes generator rotational speed data and impeller rotational speed data; the power data includes output power data and input power data; the temperature data includes gearbox oil temperature data, generator stator temperature data, and bearing temperature data; and the vibration data includes gearbox vibration acceleration data and bearing vibration acceleration data.

[0008] A further improvement of the present invention is that the step of extracting features from the data in the processing dataset using the sliding window method in step S2 includes: Based on the type of the running data, windows of different sizes are dynamically set, and multiple time-series features are extracted from each window to obtain a time-series feature set; wherein, the time-series features include the mean, variance, peak value, valley value, and trend slope of the data within the window; Based on the operational data and the environmental data, the mutual information values ​​between various parameters of the operational data and the environmental data are calculated by estimating the probability density function using a histogram. Filter parameter combinations whose mutual information values ​​are greater than a first threshold, and construct a basic association feature library based on the parameter combinations; The correlation coefficient between different parameters of the same component of the target wind turbine is calculated using the Pearson correlation coefficient. Strongly correlated feature combinations with an absolute value of the correlation coefficient greater than a second threshold are selected to form a component-level associated feature subset. The basic associated feature library and the component-level associated feature subset are merged and duplicate features are removed to obtain the associated feature set.

[0009] A further improvement of the present invention is that the step S2, which involves filtering key features using a random forest algorithm to obtain a key feature set, includes: Initialize the random forest model and set its model parameters; wherein the model parameters include the number of decision trees, the maximum depth of the trees, the minimum number of sample splits, and the minimum number of sample leaf nodes; The temporal feature set and the associated feature set are merged into an initial feature set; The initial feature set is input into the random forest model, and the amount by which each feature in the initial feature set reduces the node impurity in all decision trees is calculated to obtain the importance score of each feature. Sort all features in descending order of importance score, calculate the cumulative percentage of the importance score, and select the top N features whose cumulative percentage reaches a third threshold to obtain the key feature set.

[0010] A further improvement of the present invention is that step S4 specifically includes: Step S41: Construct and train the model based on the LSTM network to obtain the LSTM sub-model; Step S42: Build and train a model based on XGBoost technology to obtain an XGBoost sub-model; Step S43: Convert the training set into a first training dataset corresponding to the LSTM sub-model data format and a second training dataset corresponding to the XGBoost sub-model data format; Step S44: Input the first training dataset and the second training dataset into the LSTM sub-model and the XGBoost sub-model respectively, and output the first prediction result and the second prediction result respectively; Step S45: Introduce an attention mechanism to assign weights to the first prediction result and the second prediction result, and obtain the prediction result of the running state prediction model by weighted summation based on the assignment result; Step S46: Encapsulate the LSTM sub-model, the XGBoost sub-model, and the results of weight allocation and weighted calculation into the running state prediction model; The operation status prediction model takes the operation data and environmental data of the target wind turbine as input and the operation status prediction result of the target wind turbine as output.

[0011] A further improvement of the present invention is that step S45 further includes: taking The validation set is converted into a first validation dataset corresponding to the LSTM sub-model data format and a second validation dataset corresponding to the XGBoost sub-model data format; The first validation dataset is input into the LSTM sub-model, and the first mean absolute error of the LSTM sub-model on the first validation dataset is extracted. The second validation dataset is input into the XGBoost sub-model, and the second mean absolute error of the XGBoost sub-model on the second validation dataset is extracted. The weight allocation is performed based on the first mean absolute error and the second mean absolute error through the attention mechanism.

[0012] A further improvement of this invention is that the calculation formula for the weight allocation is: ; ; In the formula, This represents the weights of the LSTM sub-model. This represents the weights of the XGBoost sub-model. This represents the first mean absolute error. This represents the second mean absolute error.

[0013] A system for predicting the operating status of a wind turbine generator, comprising: The data acquisition module is used to collect the operating data and environmental data of the target wind turbine, and to preprocess the operating data and environmental data to obtain a processed dataset. The feature extraction and filtering module is used to extract features from the data in the processing dataset using the sliding window method, and to filter key features using the random forest algorithm to obtain a key feature set. The dataset partitioning module is used to divide the key feature set into training set, validation set and test set according to a certain ratio; The model building and training module is used to build a fusion prediction model based on the fusion of LSTM network and XGBoost. The training set is input into the fusion prediction model for model training, and the training parameters of the fusion prediction model are improved based on the training results to obtain the running state prediction model. The prediction result generation module is used to input the test set into the operation status prediction model to obtain the operation status prediction result of the target wind turbine.

[0014] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for predicting the operating state of a wind turbine generator.

[0015] Compared with the prior art, the present invention has at least the following beneficial technical effects: This invention provides a method and system for predicting the operating status of wind turbine generators. The invention comprehensively covers key factors affecting the operating status of wind turbine generators by collecting operating data and environmental data from multiple dimensions, providing a sufficient data foundation for subsequent predictions and avoiding the bias caused by single data dimensions. Data preprocessing eliminates noise and outliers, effectively improving data quality and reducing the interference of noise and outliers on the model's prediction accuracy. Key features are then extracted using the sliding window method and random forest algorithm, preserving key information while reducing model training complexity. The constructed improved fusion prediction model combines the fitting advantages of LSTM for time-series data with the ability of XGBoost to capture nonlinear relationships, improving prediction accuracy and achieving precise prediction of the operating status of wind turbine generators. This invention effectively improves the accuracy and real-time performance of wind turbine generator operating status prediction, providing a scientific basis for wind turbine generator operation and maintenance, reducing generator downtime, lowering operation and maintenance costs, and ensuring the stable operation of wind farms. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the prediction method based on the operating status of wind turbine units according to Embodiment 1 of the present invention. Figure 2 This is a step-by-step flowchart of step S4 of the prediction method in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the module of the prediction system based on the operating status of wind turbine units according to Embodiment 2 of the present invention. Detailed Implementation

[0018] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0019] In the description of this invention, it should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0020] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0022] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0023] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] Example 1 like Figure 1 As shown in the figure, this embodiment provides a method for predicting the operating status of a wind turbine generator set. The prediction method includes: Step S1: Collect the operating data and environmental data of the target wind turbine, and preprocess the operating data and environmental data to obtain the processed dataset.

[0025] In this embodiment, the operational data includes the target wind turbine's rotational speed, power, temperature, vibration, and voltage / current data; the environmental data includes wind speed, wind direction, ambient temperature, ambient humidity, and atmospheric pressure data; specifically, the rotational speed data includes generator rotational speed and rotor rotational speed; the power data includes output power and input power; the temperature data includes gearbox oil temperature, generator stator temperature, and bearing temperature; and the vibration data includes gearbox vibration acceleration and bearing vibration acceleration. This embodiment collects wind turbine operational and environmental data from multiple dimensions, with a data acquisition frequency of 1-5Hz to ensure the real-time and continuous nature of the collected data, thereby enabling accurate prediction of the wind turbine's operating status.

[0026] As an optional embodiment, the step S1 of preprocessing the running data and environmental data to obtain the processed dataset includes: performing noise removal, outlier handling, and data normalization on the running data and environmental data; specifically, firstly, high-frequency noise in the running data and environmental data is removed using a wavelet transform algorithm; secondly, outliers in the running data and environmental data are identified based on the 3σ criterion. If the absolute value of the difference between a data point and the mean of the data sequence is greater than 3 times the standard deviation, it is determined to be an outlier; linear interpolation is used to complete the outliers to ensure the integrity of the data sequence; finally, Min-Max normalization is used to map the running data and environmental data after noise removal and outlier handling to the [0,1] interval for data normalization, avoiding the impact of data magnitude differences on subsequent model training. The normalization formula is: ; In the formula, x represents the original runtime data and environment data. The minimum value in the data sequence. The maximum value in the data sequence. This is the normalized data.

[0027] Step S2: Extract features from the data in the processing dataset using the sliding window method, and filter key features using the random forest algorithm to obtain a key feature set.

[0028] In this embodiment, the step S2 of extracting features from the data in the processing dataset using the sliding window method includes: dynamically setting windows of different sizes of data points based on the type of running data, and extracting multiple time-series features within each window to obtain a set of time-series features; wherein, the time-series features include the mean, variance, peak value, valley value, and trend slope of the data within the window; for example, a sliding window with a window size of 10 to 30 data points and a step size of 5 to 10 data points can be set.

[0029] Based on operational and environmental data, the mutual information values ​​between various parameters of the operational and environmental data are calculated by estimating the probability density function using histograms; parameter combinations with mutual information values ​​greater than a first threshold are selected, and a basic association feature library is constructed based on these parameter combinations; for example, the first threshold can be set to 0.3.

[0030] The correlation coefficient between different parameters of the same component of the target wind turbine is calculated using the Pearson correlation coefficient. Strongly correlated feature combinations with an absolute value of correlation coefficient greater than a second threshold are selected to form a component-level associated feature subset; for example, the second threshold can be set to 0.6.

[0031] The basic associated feature library and the component-level associated feature subset are merged and duplicate features are removed to obtain the associated feature set.

[0032] As an optional embodiment, step S2, which involves selecting key features using a random forest algorithm to obtain a key feature set, includes: first, initializing the random forest model and setting its parameters; these parameters include the number of decision trees, the maximum tree depth, the minimum number of sample splits, and the minimum number of sample leaf nodes. For example, setting the number of decision trees to 100, the maximum tree depth to 8, the minimum number of sample splits to 2, and the minimum number of sample leaf nodes to 1 ensures that the random forest model can fully learn feature importance while avoiding overfitting. Second, merging the temporal feature set and the associated feature set into an initial feature set. Next, inputting the initial feature set into the random forest model, calculating the reduction in node impurity for each feature in the initial feature set across all decision trees, and obtaining the importance score for each feature. Finally, sorting all features in descending order of importance score and calculating the cumulative percentage of importance scores, selecting the top N features with a cumulative percentage reaching a third threshold to obtain the key feature set. For example, sorting by feature importance and selecting the top 80% of features to form the key feature set reduces the interference of redundant features on model training.

[0033] Step S3: Divide the key feature set into training set, validation set and test set according to the proportion.

[0034] In this embodiment, the key feature set is randomly divided into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0035] Step S4: Construct a fusion prediction model based on the combination of LSTM network and XGBoost technology. Input the training set into the fusion prediction model for model training, and improve the training parameters of the fusion prediction model based on the training results to obtain the running state prediction model.

[0036] In this embodiment, as Figure 2 As shown, step S4 specifically includes: Step S41: Construct and train an LSTM (Long Short-Term Memory) network to obtain an LSTM sub-model. Specifically, the core advantage of LSTM is its ability to process time-series data. Wind turbine data changes over time; for example, rotational speed and temperature change every second, and the data from different times are correlated. Therefore, the LSTM sub-model is used to capture the trend of data changes over time. For example, the gearbox vibration has gradually increased over the past 10 minutes, which may lead to gearbox malfunctions in the next hour.

[0037] In this embodiment, the LSTM network in the LSTM sub-model has 2-3 layers, with 64-128 hidden units per layer. The activation function is ReLU, the optimizer is Adam, and the loss function is mean squared error (MSE).

[0038] Step S42: Construct and train a model based on XGBoost (Extreme Gradient Boosting Tree) technology to obtain an XGBoost sub-model. Specifically, the core advantage of XGBoost is its ability to handle nonlinear relationships, as the operating state of wind turbines is affected by a combination of multiple factors. Therefore, the XGBoost sub-model is used to capture the complex correlations between different feature combinations. For example, when the wind speed is >15m / s and the gearbox oil temperature is >75℃, the probability of a decrease in the output power of the wind turbine increases.

[0039] In this embodiment, the XGBoost sub-model is constructed by setting the number of decision trees to 100-200, the maximum tree depth to 7, and the learning rate to 0.07.

[0040] Step S43: Convert the training set into a first training dataset corresponding to the LSTM sub-model data format and a second training dataset corresponding to the XGBoost sub-model data format.

[0041] Step S44: Input the first training dataset and the second training dataset into the LSTM sub-model and the XGBoost sub-model respectively, and output the first prediction result and the second prediction result respectively.

[0042] In this embodiment, an LSTM sub-model is trained using a first training dataset, and the training parameters of the LSTM sub-model are adjusted using a validation set to enable the LSTM sub-model to accurately fit the temporal variation trend of the wind turbine's operating state. An XGBoost sub-model is trained using a second training dataset, and the model hyperparameters of the XGBoost sub-model are optimized using a validation set to enable the XGBoost sub-model to effectively capture the nonlinear relationships between features.

[0043] Step S45: Introduce an attention mechanism to assign weights to the first and second prediction results, and then sum the weighted results to obtain the prediction results of the running state prediction model.

[0044] As an optional embodiment, step S45 further includes: converting the validation set into a first validation dataset corresponding to the LSTM sub-model data format and a second validation dataset corresponding to the XGBoost sub-model data format. The first validation dataset is input into the LSTM sub-model, and a first mean absolute error (MAE) of the LSTM sub-model on the first validation dataset is extracted. The second validation dataset is input into the XGBoost sub-model, and a second MAE of the XGBoost sub-model on the second validation dataset is extracted. Weights are assigned based on the first and second MAEs using an attention mechanism. Specifically, an attention mechanism is introduced to assign weights to the outputs of the LSTM and XGBoost sub-models; the prediction errors of the two sub-models on the validation set are calculated, with smaller errors resulting in larger weights.

[0045] As an optional embodiment, the formula for calculating the weight allocation is: ; ; In the formula, This represents the weights of the LSTM sub-model. This represents the weights of the XGBoost sub-model. Indicates the first mean absolute error. This represents the second mean absolute error.

[0046] As an optional embodiment, the prediction result of the running state prediction model is obtained by weighted summation, and the calculation formula is as follows: ; In the formula, This represents the prediction results of the operational status prediction model. This indicates the first prediction result. represents the second prediction result, represents the weight of the LSTM sub-model, and represents the weight of the XGBoost sub-model.

[0047] Step S46: Encapsulate the LSTM sub-model, XGBoost sub-model, and the results of weight allocation and weighted calculation into an operating state prediction model; wherein, the operating state prediction model takes the operating data and environmental data of the target wind turbine as input and the operating state prediction result of the target wind turbine as output.

[0048] Step S5: Input the test set into the operation status prediction model to obtain the target wind turbine operation status prediction result. In this embodiment, after inputting the test set into the operation status prediction model to obtain the target wind turbine operation status prediction result, it is determined whether the predicted value is within the preset normal threshold range. If the predicted value exceeds the threshold, an early warning mechanism is triggered, and an early warning information is sent to the operation and maintenance management platform. Specifically, the early warning mechanism sends the early warning information to the operation and maintenance management platform through audible and visual alarms and remote communication, and outputs abnormal parameter information and possible fault cause analysis.

[0049] This embodiment provides a method for predicting the operating status of wind turbine generators. This method collects wind turbine operating data and environmental data from multiple dimensions, comprehensively covering key factors affecting the generator's operating status, providing a sufficient data foundation for subsequent predictions and avoiding the bias caused by a single data dimension. Data preprocessing eliminates noise and outliers, effectively improving data quality and reducing the interference of noise and outliers on the model's prediction accuracy. Then, the sliding window method and random forest algorithm are used to extract key features, preserving crucial data information while reducing model training complexity. The constructed improved fusion prediction model combines the fitting advantages of LSTM for time-series data with the ability of XGBoost to capture nonlinear relationships. Simultaneously, an attention mechanism is used to dynamically optimize the weights of sub-models, improving prediction accuracy and achieving precise prediction of the wind turbine generator's operating status. This invention effectively improves the accuracy and real-time performance of wind turbine generator operating status prediction, providing a scientific basis for wind turbine operation and maintenance, reducing generator downtime, lowering maintenance costs, and ensuring the stable operation of wind farms.

[0050] Example 2 like Figure 3 As shown, this embodiment provides a wind turbine operating status prediction system 10, which includes: a data acquisition module 11, a feature extraction and filtering module 12, a dataset partitioning module 13, a model building and training module 14, and a prediction result generation module 15.

[0051] The data acquisition module 11 is used to collect the operating data and environmental data of the target wind turbine, and to preprocess the operating data and environmental data to obtain the processed dataset.

[0052] The feature extraction and filtering module 12 is used to extract features from the data in the processing dataset using the sliding window method and to filter key features using the random forest algorithm to obtain a key feature set.

[0053] The dataset partitioning module 13 is used to divide the key feature set into training set, validation set and test set according to the proportion.

[0054] The model building and training module 14 is used to build a fusion prediction model based on the fusion of LSTM network and XGBoost. The training set is input into the fusion prediction model for model training, and the training parameters of the fusion prediction model are improved based on the training results to obtain the running state prediction model.

[0055] The prediction result generation module 15 is used to input the test set into the operation status prediction model to obtain the anomaly prediction result of the target wind turbine.

[0056] This embodiment provides a wind turbine operating status prediction system. The prediction system uses the wind turbine operating status prediction method provided in Embodiment 1 above; specific details are not repeated here. This embodiment's wind turbine operating status prediction system collects wind turbine operating data and environmental data from multiple dimensions, comprehensively covering key factors affecting the turbine's operating status, providing a sufficient data foundation for subsequent predictions and avoiding the bias caused by a single data dimension. Data preprocessing eliminates noise and outliers, effectively improving data quality and reducing the interference of noise and outliers on the model's prediction accuracy. Key features are extracted using the sliding window method and random forest algorithm, preserving key information while reducing model training complexity. The constructed improved fusion prediction model combines the fitting advantages of LSTM for time-series data with the ability of XGBoost to capture nonlinear relationships, and uses an attention mechanism to dynamically optimize sub-model weights, improving prediction accuracy and achieving precise prediction of wind turbine operating status. This invention effectively improves the accuracy and real-time performance of wind turbine operating status prediction, providing a scientific basis for wind turbine operation and maintenance, reducing turbine downtime, lowering maintenance costs, and ensuring the stable operation of wind farms.

[0057] Example 3 This embodiment relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the wind turbine operating state prediction method of Embodiment 1 described above.

[0058] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0059] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0060] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for predicting the operating status of a wind turbine generator set, characterized in that, include: Step S1: Collect the operating data and environmental data of the target wind turbine, and preprocess the operating data and environmental data to obtain a processed dataset; the operating data includes the speed data, power data, temperature data, vibration data, voltage and current data of the target wind turbine; the environmental data includes wind speed data, wind direction data, ambient temperature data, ambient humidity data, and atmospheric pressure data. Step S2: Extract features from the data in the processing dataset using the sliding window method, and filter key features using the random forest algorithm to obtain a key feature set; Step S3: Divide the key feature set into a training set, a validation set, and a test set according to the proportions; Step S4: Construct a fusion prediction model based on the fusion of LSTM network and XGBoost technology. Input the training set into the fusion prediction model for model training, and improve the training parameters of the fusion prediction model based on the training results to obtain the running state prediction model. Step S5: Input the test set into the operation status prediction model to obtain the operation status prediction result of the target wind turbine.

2. The method for predicting the operating status of a wind turbine generator according to claim 1, characterized in that, The step S1 of preprocessing the running data and the environmental data to obtain the processed dataset includes: The operating data and the environmental data are subjected to noise removal, outlier handling, and data normalization. The noise removal operation uses a wavelet transform algorithm to remove high-frequency noise from the running data and the environmental data. The outlier handling is based on the 3σ criterion to identify outliers in the operational data and the environmental data; The data normalization process uses Min-Max normalization to map the running data and the environmental data after the noise removal operation and the outlier processing to the [0,1] interval for data normalization.

3. The method for predicting the operating status of a wind turbine generator according to claim 1, characterized in that, The rotational speed data includes generator rotational speed data and impeller rotational speed data; the power data includes output power data and input power data; the temperature data includes gearbox oil temperature data, generator stator temperature data, and bearing temperature data; the vibration data includes gearbox vibration acceleration data and bearing vibration acceleration data.

4. The method for predicting the operating status of a wind turbine generator according to claim 1, characterized in that, The step of extracting features from the data in the processing dataset using the sliding window method in step S2 includes: Based on the type of the running data, windows of different sizes are dynamically set, and multiple time-series features are extracted from each window to obtain a time-series feature set; wherein, the time-series features include the mean, variance, peak value, valley value, and trend slope of the data within the window; Based on the operational data and the environmental data, the mutual information values ​​between various parameters of the operational data and the environmental data are calculated by estimating the probability density function using a histogram. Filter parameter combinations whose mutual information values ​​are greater than a first threshold, and construct a basic association feature library based on the parameter combinations; The correlation coefficient between different parameters of the same component of the target wind turbine is calculated using the Pearson correlation coefficient. Strongly correlated feature combinations with an absolute value of the correlation coefficient greater than a second threshold are selected to form a component-level associated feature subset. The basic associated feature library and the component-level associated feature subset are merged and duplicate features are removed to obtain the associated feature set.

5. The method for predicting the operating status of a wind turbine generator according to claim 4, characterized in that, The step S2, which involves filtering key features using the random forest algorithm to obtain a key feature set, includes: Initialize the random forest model and set its model parameters; wherein the model parameters include the number of decision trees, the maximum depth of the trees, the minimum number of sample splits, and the minimum number of sample leaf nodes; The temporal feature set and the associated feature set are merged into an initial feature set; The initial feature set is input into the random forest model, and the amount by which each feature in the initial feature set reduces the node impurity in all decision trees is calculated to obtain the importance score of each feature. Sort all features in descending order of importance score, calculate the cumulative percentage of the importance score, and select the top N features whose cumulative percentage reaches a third threshold to obtain the key feature set.

6. The method for predicting the operating status of a wind turbine generator according to claim 1, characterized in that, The specific steps of step S4 include: Step S41: Construct and train the model based on the LSTM network to obtain the LSTM sub-model; Step S42: Build and train a model based on XGBoost technology to obtain an XGBoost sub-model; Step S43: Convert the training set into a first training dataset corresponding to the LSTM sub-model data format and a second training dataset corresponding to the XGBoost sub-model data format; Step S44: Input the first training dataset and the second training dataset into the LSTM sub-model and the XGBoost sub-model respectively, and output the first prediction result and the second prediction result respectively; Step S45: Introduce an attention mechanism to assign weights to the first prediction result and the second prediction result, and obtain the prediction result of the running state prediction model by weighted summation based on the assignment result; Step S46: Encapsulate the LSTM sub-model, the XGBoost sub-model, and the results of weight allocation and weighted calculation into the running state prediction model; The operation status prediction model takes the operation data and environmental data of the target wind turbine as input and the operation status prediction result of the target wind turbine as output.

7. The method for predicting the operating status of a wind turbine generator according to claim 6, characterized in that, The step S45 further includes: […]. The validation set is converted into a first validation dataset corresponding to the LSTM sub-model data format and a second validation dataset corresponding to the XGBoost sub-model data format; The first validation dataset is input into the LSTM sub-model, and the first mean absolute error of the LSTM sub-model on the first validation dataset is extracted. The second validation dataset is input into the XGBoost sub-model, and the second mean absolute error of the XGBoost sub-model on the second validation dataset is extracted. The weight allocation is performed based on the first mean absolute error and the second mean absolute error through the attention mechanism.

8. The method for predicting the operating status of a wind turbine generator according to claim 7, characterized in that, The formula for calculating the weight allocation is as follows: ; ; In the formula, This represents the weights of the LSTM sub-model. This represents the weights of the XGBoost sub-model. This represents the first mean absolute error. This represents the second mean absolute error.

9. A prediction system for the operating status of a wind turbine generator set, characterized in that, include: The data acquisition module is used to collect the operating data and environmental data of the target wind turbine, and to preprocess the operating data and environmental data to obtain a processed dataset. The feature extraction and filtering module is used to extract features from the data in the processing dataset using the sliding window method, and to filter key features using the random forest algorithm to obtain a key feature set. The dataset partitioning module is used to divide the key feature set into training set, validation set and test set according to a certain ratio; The model building and training module is used to build a fusion prediction model based on the fusion of LSTM network and XGBoost. The training set is input into the fusion prediction model for model training, and the training parameters of the fusion prediction model are improved based on the training results to obtain the running state prediction model. The prediction result generation module is used to input the test set into the operation status prediction model to obtain the operation status prediction result of the target wind turbine.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for predicting the operating status of the wind turbine generator as described in any one of claims 1 to 8.

Images