A wind turbine operating state evaluation method and device

Abstract

The present application belongs to the technical field of wind turbine state evaluation, and particularly relates to a wind turbine operation state evaluation method and device. In the present application, when processing the historical monitoring data set of each working condition parameter, an improved fuzzy C-means clustering algorithm is used for clustering working condition division. The improvement of the improved fuzzy C-means clustering algorithm is that the initial clustering center is no longer determined randomly, but the density of each sampling point is used to select the initial clustering center. The density of the selected initial clustering center is larger, and the distance between each initial clustering center is relatively far, which prevents the situation that multiple clustering segmentation blocks appear in the area with small density caused by random selection of the initial clustering center. Therefore, a more suitable initial clustering center is selected, so that the label added to the training data of the working condition recognition model is accurate, and the prediction accuracy of the working condition recognition model is ensured, thereby laying a foundation for the accuracy of subsequent wind turbine operation state evaluation.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine condition assessment technology, specifically relating to a method and apparatus for assessing the operating condition of wind turbines. Background Technology

[0002] In recent years, with many countries increasing their investment in wind power technology research, the wind power industry has experienced rapid development both domestically and internationally, and the global wind power market is expanding rapidly. However, wind turbines are typically located in areas rich in wind resources, such as high mountains, deserts, and offshore islands, where the operating environment is harsh and the operating conditions are complex and variable, leading to frequent turbine failures and high annual operating and maintenance costs. Therefore, conducting research on the operational health status assessment of wind turbines, timely and accurately grasping the operating conditions of the units, and achieving early warning of failures are of great guiding significance for improving the safe and reliable operation of wind power systems and reducing operating and maintenance costs.

[0003] Currently, wind turbines mainly rely on traditional condition monitoring systems for operation and maintenance, which suffer from inaccurate and untimely fault alarms. When the system issues a fault alarm, the current fault may have become so severe that it is necessary to "shut down" or even "emergency shutdown," leaving no time to prevent the unit fault from worsening. Moreover, the massive amount of wind turbine operation data stored in the existing wind farm big data system is not being fully utilized.

[0004] To address the aforementioned issues, Chinese invention patent application CN111709490A discloses a wind turbine health status assessment method based on a GRU neural network. This method first inputs real-time wind turbine feature parameters into a random forest classification model. Then, based on the classification results, it maps the feature parameter data to corresponding operating conditions for condition matching. Next, it obtains the predicted values ​​of the real-time sample data from the GRU neural network evaluation model for the corresponding operating condition. The average absolute percentage error between the predicted values ​​and monitored values ​​from the SCADA system is calculated to obtain the wind turbine operating health index. Finally, the health index is used to assess the wind turbine's operating health status. The prerequisite for predicting the SCADA system's predicted values ​​in this method is determining which operating condition the real-time wind turbine feature parameters correspond to. Incorrect determination of the operating condition will lead to the subsequent incorrect selection of the GRU neural network evaluation model. To accurately determine the operating condition, the random forest classification model requires high accuracy, and the labels of its training data are crucial for ensuring accuracy. This method utilizes the CLIQUE clustering algorithm for operating condition classification, which can improve operational efficiency and accuracy to a certain extent. Of course, other clustering algorithms can also be chosen for operating condition classification, such as the fuzzy C-means clustering algorithm (FCM). However, FCM clustering involves a degree of randomness in selecting initial cluster centers, especially when clustering large datasets. Using poor initial cluster centers can have a greater impact on the clustering results. Currently, most methods for selecting initial cluster centers are random, affecting clustering accuracy. This compromised accuracy inevitably leads to inaccurate final health status assessments of wind turbines. Summary of the Invention

[0005] The purpose of this invention is to provide a method and apparatus for evaluating the operating status of wind turbine units, so as to solve the problem of inaccurate evaluation of the operating health status of wind turbine units in the prior art.

[0006] To address the aforementioned technical problems, this invention provides a method for evaluating the operating status of wind turbine generators, comprising the following steps:

[0007] 1) Obtain historical monitoring datasets for various operating parameters of the wind turbine generator set. Extract features from each set of historical monitoring data in the datasets for each operating condition parameter. Use the extracted features to cluster each set of historical monitoring data using an improved fuzzy C-means clustering algorithm. Use the clustering results as the operating condition classification results for each set of historical monitoring data. The improvement of the improved fuzzy C-means clustering algorithm lies in determining the initial cluster centers using the following method:

[0008] ① Calculate x for each sample point i Density function values, i = 1, 2, ..., n, where n represents the total number of sample points: Where x jLet || denote the j-th sample point, and || denote the Euclidean distance. Select the K sample points with larger density function values ​​from them to form a filter set, where K < n;

[0009] ② Select the two farthest sample points from the filter set as initial cluster centers. Determine whether the number of initial cluster centers selected at this time is greater than or equal to the set number of cluster centers required. If it is greater than or equal to, then the finally selected sample points are used as initial cluster centers. Otherwise, the sample points selected as initial cluster centers are deleted from the filter set, and step ② is repeated on the filter set after the deletion operation until the number of selected cluster centers is greater than or equal to the set number of cluster centers required.

[0010] 2) Use the historical monitoring dataset of each working condition parameter with known working condition division results to train the constructed working condition identification model to obtain the trained working condition identification model; obtain the monitoring data of each working condition parameter of the wind turbine during the evaluation period and input it into the trained working condition identification model to obtain the working condition of the wind turbine during the evaluation period.

[0011] 3) Obtain the monitoring data of each state parameter of the wind turbine during the evaluation period, and input them into the state parameter prediction model corresponding to the operating condition and each state parameter to obtain the prediction data of each state parameter of the wind turbine during the prediction period.

[0012] 4) Determine the difference between the predicted data and the monitoring data of each state parameter of the wind turbine during the prediction period, and use the difference to evaluate the operating status of the wind turbine.

[0013] The beneficial effects are as follows: When processing condition labels on historical monitoring datasets of various operating condition parameters, this invention employs an improved fuzzy C-means clustering algorithm for clustering operating conditions. The improvement lies in no longer randomly determining initial cluster centers. Instead, it selects initial cluster centers based on the density of each sampling point. The selected initial cluster centers have a higher density and are relatively far apart, preventing the occurrence of multiple clustering blocks in low-density areas due to random selection. This results in selecting more suitable initial cluster centers. Using appropriate initial cluster centers not only avoids the impact of outliers and isolated points on clustering but also reduces running time, improves computational efficiency, saves computing resources, and enhances clustering accuracy. This ensures that the labels added to the training data of the operating condition identification model are accurate, thereby guaranteeing the prediction accuracy of the model and laying the foundation for the accuracy of subsequent wind turbine operating status assessments. This invention can sensitively track the deterioration process of wind turbines and accurately assess changes in the health status of wind turbine operation, which has important guiding significance for reducing the operation and maintenance costs of wind turbines and ensuring the safe and reliable operation of wind power systems.

[0014] Furthermore, the number of clusters in the improved fuzzy C-means clustering algorithm is:

[0015]

[0016]

[0017]

[0018] Among them, c * U represents the final number of clusters selected; * Ω represents the membership matrix at the optimal number of clusters; U represents the membership matrix consisting of memberships u; c R represents the optimal finite set; ik The fuzzy correlation between class i and class k is represented by m; m represents the weight; n is the total number of sample points; ||x j -v i ||and||x j -v k || represents the sample point x j Each with cluster center v i v k Euclidean distance between them; u ji and u jk Represents sample point x j The degree of membership in the i-th and k-th classes, respectively, and

[0019] Its beneficial effects are: by using the fuzzy correlation function to determine the optimal number of clusters, clusters with high fuzzy correlation can be appropriately aggregated into one cluster, thereby making the fuzzy correlation between the final clusters smaller.

[0020] Further, the operating parameters include wind speed and power; the historical monitoring dataset of each operating parameter in step 1) is the historical monitoring dataset of each operating parameter after preprocessing the historical original monitoring dataset of each operating parameter. The preprocessing method includes: using the power of the wind turbine and the wind speed of the wind turbine to cluster the historical original monitoring dataset of each operating parameter using the CLIQUE clustering algorithm, and removing outlier historical original monitoring data according to the clustering results.

[0021] Its beneficial effect is that it first uses wind speed and power to eliminate obviously problematic data, thus ensuring the accuracy of subsequent calculations and processing.

[0022] Furthermore, the operating parameters are the effective operating parameters selected from the candidate operating parameters. The selected effective operating parameters include wind speed, power, generator speed, impeller speed, torque, and ambient temperature.

[0023] Furthermore, the working condition identification model in step 2) is a random forest model; the state parameter prediction model in step 3) is a BiGRU neural network model.

[0024] Its beneficial effects are: choosing the random forest model can improve classification accuracy; compared with the traditional GRU neural network, the chosen BiGRU neural network model can extract valuable information from massive amounts of data more effectively.

[0025] Furthermore, the state parameters are valid state parameters selected from the candidate state parameters. The selection method is as follows: the FP-Growth algorithm is used to mine the correlation between the candidate state parameters and the degradation mode of related components, and the state parameters with a correlation greater than a set threshold are selected as valid state parameters.

[0026] Its beneficial effects are as follows: the FP-Growth algorithm is used to select effective state parameters from the candidate state parameters. On the one hand, it avoids the problem of large amount of computation caused by using too many state parameters. On the other hand, selecting more effective state parameters can improve the evaluation accuracy.

[0027] Further, in step 4), the method for evaluating the operating status of the wind turbine using the aforementioned differences is as follows: The health index value of the wind turbine is calculated using the differences, and the operating status of the wind turbine is evaluated based on the health index value. The health index value is:

[0028]

[0029]

[0030] Among them, HI (t) This represents the health index value of the wind turbine at time t; This represents the weight of the i-th state parameter at time t; n represents the total number of state parameters. represents the mean error of the i-th state parameter within a sliding window; m represents the time length of the sliding window. Let represent the deviation between the predicted data and the monitored data of the i-th state parameter at time t, and

[0031]

[0032] in, This represents the predicted data for the i-th state parameter at time t; σ represents the monitoring data of the i-th state parameter at time t; σ represents a real number set to prevent the denominator from being 0.

[0033] Its beneficial effects are: it utilizes real-time health index values ​​of wind turbines to achieve quantitative assessment of their health status and early fault warning; moreover, in calculation... Adding a real number σ to the denominator can prevent the denominator from being zero and thus making calculation impossible. This situation has occurred.

[0034] Furthermore, the weight of the i-th state parameter at time t The update formula is:

[0035]

[0036] in, This represents the weight of the i-th state parameter at the initial time.

[0037] Its beneficial effect is: utilizing the mean error within the sliding window To calculate weights Compared to using Calculate weights In order to prevent a certain moment The mutation caused the weight The problem of mutations affecting the accuracy of health index calculations.

[0038] Furthermore, the method for evaluating the operating status of wind turbine units based on health index values ​​is as follows: determining the health index range in which the health index value falls, and determining the status level of the wind turbine unit based on the health index range in which it falls; wherein, one health index range corresponds to one status level, and there are a total of 5 health index ranges.

[0039] Its beneficial effects are: dividing the wind turbine into 5 health index intervals allows for a precise classification of the wind turbine's operating status.

[0040] To address the aforementioned technical problems, the present invention also provides a wind turbine operating status assessment device. The assessment device includes a processor, which executes computer program instructions to implement the wind turbine operating status assessment method described above, and achieves the same beneficial effects as the method. Attached Figure Description

[0041] Figure 1 This is a flowchart of the working condition identification process of the present invention;

[0042] Figure 2 This is a flowchart of the wind turbine operating status assessment method of the present invention;

[0043] Figure 3 This is a schematic diagram of the BiGRU neural network structure used in this invention;

[0044] Figure 4 This is a structural diagram of the wind turbine operation status assessment device of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, not all embodiments. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of this invention.

[0046] Example of wind turbine operating status assessment method:

[0047] An embodiment of the wind turbine operating status assessment method of the present invention, the overall process of which is as follows: Figure 2 As shown, the process is as follows:

[0048] Step one: Build and train a random forest model to obtain a working condition identification model. The entire process is as follows:

[0049] 1. Obtain historical monitoring datasets of potential operating parameters for the wind turbine from the SCADA system. These parameters include wind speed, power, generator speed, rotor speed, torque, ambient temperature, gearbox oil temperature, and average pitch angle. The historical monitoring dataset consists of multiple sets of historical monitoring data. Each set contains the specific values ​​of these parameters at a given moment.

[0050] 2. Use wind speed and power from the candidate operating condition parameters to preprocess multiple sets of historical monitoring data in the historical monitoring dataset to remove problematic data. The specific processing method is as follows: use the CLIQUE clustering algorithm to cluster the historical monitoring dataset of operating condition parameters using wind speed and power, and remove outlier historical monitoring data based on the clustering results, that is, remove data with obvious problems in power and wind speed.

[0051] 3. Perform correlation analysis on the historical monitoring datasets of the selected operating parameters to screen out the effective operating parameters. The screened effective operating parameters include: wind speed, power, generator speed, impeller speed, torque, and ambient temperature.

[0052] 4. The KPCA (Kernel Principal Component Analysis) algorithm is used to extract features from the historical monitoring dataset of the selected effective operating parameters. The feature extraction results are then used as input to the improved FCM clustering algorithm for cluster analysis. The resulting clusters are used as the operating condition classification for each group of historical monitoring data. Each cluster represents one operating condition of the wind turbine, and operating condition labels are assigned to each group of historical monitoring data for effective operating parameters in this manner. The improved FCM clustering algorithm used in this embodiment includes two improvements: one is the method for selecting initial cluster centers, and the other is the method for selecting the number of clusters. These are described in detail below.

[0053] 1) The initial cluster centers of the FCM clustering algorithm are determined based on the density function. The process of determining the initial cluster centers is as follows:

[0054] ① Given a sample dataset X = {x1, x2, ..., x...} n}, x i Let i be the i-th sample point, i = 1, 2, ..., n, where n represents the total number of sample points; set the required number of cluster centers c and the number of counts t = 0.

[0055] ② Calculate the density function value of each sample point according to the following formulas (1) and (2), sort them in descending order, and select the first n / 2 values ​​to add to set E.

[0056]

[0057] Where, ρ i Represents sample point x i Density value; n represents the total number of sample points; ||x i -x j || represents the sample point x i and sample point x j Euclidean distance; This represents the average distance between all sample points, i.e.

[0058]

[0059] ③ Select the two sample points that are farthest apart in set E as two initial cluster centers v1 and v2 respectively, and add them to set V. At the same time, delete v1 and v2 from set E.

[0060] ④ Continue to select the data point that is furthest from the existing cluster center in set V from set E as the new cluster center and add it to set V, while deleting it from set E.

[0061] ⑤ Determine if the number of cluster centers t in set V satisfies t≥c. If it does, the algorithm terminates and outputs the final initial cluster centers V = [v1, v2, ..., v c (This case is when t = c), otherwise, jump to step 4) to continue filtering until t ≥ c is satisfied.

[0062] 2) Determine the optimal number of clusters using the fuzzy correlation function. The formula for calculating the fuzzy correlation function is:

[0063]

[0064] In the formula, U represents the membership matrix composed of membership degrees u, c represents the number of clusters, and R ik Let be the fuzzy correlation between the i-th class and the k-th class. The formula for the fuzzy correlation is as follows:

[0065]

[0066] In the formula, m is the weight; n is the size of the sample data; ||x j -v i ||and||x j -v k || represent sample points x respectively j Each with cluster center v i v k Euclidean distance between them; u ji and u jk Represents sample point x j The degree of membership to the i-th and k-th classes, respectively, for example:

[0067]

[0068] The smaller the nominal fuzzy correlation between classes, the better the clustering effect. Let Ω c Denotes the finite set of "optimal", (U * c * ) represents the optimal efficient clustering, c * U represents the optimal number of clusters. * This represents the membership matrix at the optimal number of clusters. (U) * ;c * The following formula must be satisfied:

[0069]

[0070] 5. Construct a random forest algorithm model. Train the random forest model using the feature extraction results of historical monitoring datasets with operating condition labels. After training, the operating condition identification model can be obtained.

[0071] Step two involves acquiring monitoring data of the effective operating parameters of the wind turbine during the evaluation period, performing feature extraction using KPCA, and inputting the feature extraction results into the operating condition identification model to obtain the operating condition identification result, i.e., the operating condition of the wind turbine during the evaluation period. The entire process from step one to step two is as follows: Figure 1 As shown.

[0072] Step 3: Establish and train a bidirectional recurrent neural network (BiGRU) model to obtain the state parameter prediction model. In this embodiment, the number of state parameter prediction models is related to the number of effective state parameters (p1) and the number of operating conditions (p2). The number of state parameter prediction models is p1*p2, meaning that one effective state parameter under one operating condition corresponds to one state parameter prediction model. The training process for each state parameter prediction model is described in detail below:

[0073] 1. Obtain historical monitoring datasets of candidate state parameters for wind turbine units from the SCADA system. Based on wind turbine knowledge and expert experience, 34 parameters with strong correlation to the unit status were selected from the parameters of key components for wind turbine operation and maintenance as candidate state parameters, as detailed in Table 1.

[0074] Table 1

[0075]

[0076] 2. The correlation between candidate state parameters and the degradation modes of related components is mined using the FP-Growth algorithm, and candidate state parameters with a correlation greater than 0.6 are selected as effective state parameters.

[0077] 3. Based on the working condition segmentation results, the corresponding bidirectional recurrent neural network (BiGRU) models are trained using sample data of the effective state parameters corresponding to each working condition. After training, the state parameter prediction models corresponding to each effective state parameter under each working condition are obtained. BiGRU is composed of two unidirectional GRUs stacked on top of each other, and the output is jointly determined by the states of these two GRUs. This increases the number of model parameters and effectively improves the model's learning ability. Therefore, compared with traditional GRU neural networks, BiGRU neural networks can more effectively extract valuable information from massive amounts of data. Its model structure is as follows: Figure 3 As shown.

[0078] The following section introduces one effective state parameter prediction model as an example. The construction and training process of other effective state parameter prediction models are similar. For example, for an effective state parameter prediction model with the operating condition of "constant power operation" and the effective state parameter being "power," a bidirectional recurrent neural network (BiGRU) model is first constructed. Then, a power dataset under constant power operation is used. The power dataset includes multiple sets of power data, each set containing power data at different times. The BiGRU model is trained using this power dataset. After training, an effective state parameter prediction model with the operating condition of "constant power operation" and the effective state parameter being "power" is obtained. The essence of each effective state parameter prediction model lies in using historical data to predict its development trend in order to obtain data for future times. Of course, for predicting a certain effective state parameter, the input is not necessarily limited to that effective state parameter; other effective state parameters can also be involved. This is because various effective state parameters influence each other, and the changes in other effective state parameters can also reveal the changes in the effective state parameter to be predicted.

[0079] Step 4: Obtain the monitoring data of the effective state parameters of the wind turbine during the evaluation period, and input them into the state parameter prediction model corresponding to the operating condition and each effective state parameter to obtain the prediction data of each effective state parameter of the wind turbine during the prediction period.

[0080] Step 5: Calculate the difference between the predicted data and the monitored data (real-time monitoring data) of the effective state parameters of the wind turbine during the prediction period. A dynamic weighting method is used to weight and fuse the deviation values ​​of each effective state parameter within a sliding window to obtain the health index value. The specific calculation formula for the health index value is as follows:

[0081]

[0082]

[0083] In the formula, HI (t)This represents the health index value of the wind turbine at time t; This represents the weight of the i-th state parameter at time t; n represents the total number of state parameters. represents the mean error of the i-th state parameter within a sliding window; m represents the time length of the sliding window. This indicates the deviation between the predicted data and the monitored data, i.e.

[0084]

[0085] In the formula, This represents the predicted value of the i-th state parameter at time t; σ represents the actual monitored value of the i-th state parameter at time t; σ represents a set real number that is infinitely small to prevent the denominator from being zero.

[0086] The formula for updating dynamic weights is as follows:

[0087]

[0088] Step Six: Using this health index value, a segmented threshold method is employed to classify the current state of the wind turbine into five levels. Based on expert experience and knowledge, the wind turbine health status is categorized into five levels. As shown in Table 2, Level 5 represents normal wind turbine operation; Level 4 represents minor degradation but still safe operation; Level 3 represents initial degradation, requiring a maintenance warning; Level 2 represents significant degradation with frequent malfunctions, necessitating shutdown for maintenance; and Level 1 represents severe degradation, at which point the wind turbine system will automatically shut down and issue an alarm.

[0089] Table 2

[0090]

[0091] In summary, this invention first performs correlation analysis on the operating parameters of the wind turbine unit to screen out effective operating parameters, then uses the KPCA method to extract features from these parameters, and employs an improved FCM clustering algorithm to classify operating conditions. Secondly, it uses the FP-Growth algorithm to select state assessment parameters, and based on the operating condition classification results, constructs a BiGRU model for each sub-operating condition to predict the changes in each state assessment parameter. Finally, it combines dynamic weights to perform weighted analysis on the deviations between the actual monitored values ​​and the model predicted values ​​of multiple assessment parameters, obtaining real-time health index values ​​for the wind turbine, thus achieving quantitative assessment of the wind turbine's health status and early fault warning. This invention can sensitively track the wind turbine's deterioration process and accurately assess changes in the wind turbine's operational health status, providing important guidance for reducing wind turbine operation and maintenance costs and ensuring the safe and reliable operation of wind power systems.

[0092] Example of a wind turbine operating status assessment device:

[0093] An embodiment of the wind turbine operating status assessment device of the present invention, such as... Figure 4 As shown, the system includes a memory, a processor, and an internal bus. The processor and memory communicate and interact with each other via the internal bus. The memory includes at least one software function module stored in the memory. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the wind turbine operating status assessment method described in the embodiments of the present invention.

[0094] The processor can be a microprocessor (MCU), a programmable logic device (FPGA), or other processing devices. The memory can be any type of memory that stores information using electrical energy, such as RAM and ROM; it can also be any type of memory that stores information using magnetic energy, such as hard disks, floppy disks, magnetic tapes, magnetic core memory, bubble memory, and USB flash drives; it can also be any type of memory that stores information using optical methods, such as CDs and DVDs; and of course, it can also be other types of memory, such as quantum memory and graphene memory.

[0095] Specific implementation methods have been given above, but the present invention is not limited to the described implementation methods. The basic idea of ​​the present invention lies in the above basic scheme. For those skilled in the art, designing various modified models, formulas, and parameters based on the teachings of the present invention does not require creative effort. Changes, modifications, substitutions, and variations made to the implementation methods without departing from the principles and spirit of the present invention still fall within the protection scope of the present invention.

Claims

1. A method for evaluating the operating status of a wind turbine generator set, characterized in that, include: 1) Obtain historical monitoring datasets for various operating parameters of the wind turbine generator set. Extract features from each set of historical monitoring data in the datasets for each operating condition. Use the extracted features to cluster each set of historical monitoring data using an improved fuzzy C-means clustering algorithm. Use the clustering results as the operating condition classification results for each set of historical monitoring data. The improvement of the improved fuzzy C-means clustering algorithm lies in determining the initial cluster centers using the following method: ① Calculate each sample point The density function value, , Indicates the total number of sample points: , For the first One sample point, This indicates the calculation of the Euclidean distance. Select K sample points with larger density function values ​​from these samples to form a filter set, where K < 0. ; ② Select the two farthest sample points from the selected set as initial cluster centers, and determine whether the number of initial cluster centers selected at this time is not less than the set requirement of the number of cluster centers. If the number of selected sample points is not less than the initial cluster center, then the selected sample points will be used as the initial cluster centers; otherwise, the selected sample points will be removed from the filter set, and step ② will be repeated on the filter set after the deletion operation until the number of selected cluster centers is not less than the initial cluster center. ; 2) Use the historical monitoring dataset of each working condition parameter with known working condition division results to train the constructed working condition identification model to obtain the trained working condition identification model; obtain the monitoring data of each working condition parameter of the wind turbine during the evaluation period and input it into the trained working condition identification model to obtain the working condition of the wind turbine during the evaluation period. 3) Obtain the monitoring data of each state parameter of the wind turbine during the evaluation period, and input them into the state parameter prediction model corresponding to the operating condition and each state parameter to obtain the prediction data of each state parameter of the wind turbine during the prediction period. 4) Determine the differences between the predicted data and the monitoring data of each state parameter of the wind turbine during the prediction period, and use these differences to obtain the health index of the wind turbine. ,use The operating status of the wind turbine units was assessed; and ； ； ； For the first The state parameters are in The weight of each moment; The total number of state parameters; For the first The mean error of a state parameter within a sliding window; The duration of the sliding window; Indicates the first The state parameters are in Predicted data for each moment With monitoring data Deviation between; This is a real number set to prevent the denominator from being 0.

2. The wind turbine operating status assessment method according to claim 1, characterized in that, The number of clusters in the improved fuzzy C-means clustering algorithm is: ； ； ； in, This represents the final number of clusters selected; This represents the membership matrix at the optimal number of clusters. This represents the membership matrix consisting of membership degrees u; Represents the optimal finite set; Indicates the first Class and First Fuzzy relevance of classes; Indicates the weight; This represents the total number of sample points. and Represents sample points Respectively with cluster centers , The Euclidean distance between them; and Represents sample points They belong to the first Class and First The degree of membership of the class, and .

3. The wind turbine operating status assessment method according to claim 1, characterized in that, The operating parameters include wind speed and power; the historical monitoring dataset of each operating parameter in step 1) is the historical monitoring dataset of each operating parameter after preprocessing the historical original monitoring dataset of each operating parameter. The preprocessing method includes: using the power of the wind turbine and the wind speed of the wind turbine to cluster the historical original monitoring dataset of each operating parameter using the CLIQUE clustering algorithm, and removing outlier historical original monitoring data according to the clustering results.

4. The wind turbine operating status assessment method according to claim 1, characterized in that, The operating parameters are the effective operating parameters selected from the candidate operating parameters. The selected effective operating parameters include wind speed, power, generator speed, impeller speed, torque, and ambient temperature.

5. The wind turbine operating status assessment method according to claim 1, characterized in that, The working condition identification model in step 2) is a random forest model; the state parameter prediction model in step 3) is a BiGRU neural network model.

6. The wind turbine operating status assessment method according to claim 1, characterized in that, The state parameters are the valid state parameters selected from the candidate state parameters. The selection method is as follows: the FP-Growth algorithm is used to mine the correlation between the candidate state parameters and the degradation mode of related components, and the state parameters with a correlation greater than a set threshold are selected as valid state parameters.

7. The wind turbine operating status assessment method according to claim 1, characterized in that, No. The state parameters are in Time weight The update formula is: ； in, Indicates the first The weights of each state parameter at the initial time.

8. The wind turbine operating status assessment method according to claim 1, characterized in that, use The method for assessing the operating status of wind turbine units is as follows: determine the health index range in which the health index value falls, and determine the status level of the wind turbine unit based on the health index range in which it falls; wherein, one health index range corresponds to one status level, and there are a total of 5 health index ranges.

9. A wind turbine operating status assessment device, characterized in that, The evaluation device includes a processor for executing computer program instructions to implement the wind turbine operating status evaluation method as described in any one of claims 1 to 8.

Images