Health status assessment method for wind turbine generators considering operating conditions

Abstract

This invention discloses a method for assessing the health status of wind turbine generators considering operating conditions, including: preprocessing historical SCADA data of the wind turbine generators; selecting relevant feature parameters for each prediction parameter using the mRMR algorithm; classifying the operating conditions of the wind turbine generators; constructing a multi-parameter prediction model based on CNN-BiGRU for each classified operating condition; dividing the historical data of the wind turbine's healthy operating state into operating conditions, and inputting the feature parameter data of each operating condition into training and test sets into the CNN-BiGRU prediction model to obtain predicted values, and then constructing a standard residual set based on the residuals between the predicted values ​​and actual values ​​in the training set; determining the health status level index; and determining the current health level of the wind turbine generators based on real-time SCADA data. The assessment results of this invention are more objective and accurate, and can detect the deterioration trend of wind turbine generators in advance.

Description

Technical Field

[0001] This invention belongs to the technical field of wind turbine condition assessment methods, specifically relating to a method for assessing the health status of wind turbines that takes into account operating conditions. Background Technology

[0002] Timely and accurate understanding of the health status of wind turbines is crucial for optimizing wind turbine control and maintenance strategies, reducing maintenance costs, and ensuring safe and efficient operation. Currently, there are two main methods for assessing the health status of wind turbines: model-based and data-driven. Model-based methods primarily involve establishing subsystem health status models of the wind turbine and assessing the health status of these subsystems based on these models. However, this method is difficult to implement, the assessment results depend heavily on the accuracy of the established model, and because it only models subsystems, it cannot assess the overall health status of the turbine. Data-driven methods, on the other hand, are easier to implement and provide more accurate results. This invention utilizes turbine operation data from a Supervisory Control and Data Acquisition (SCADA) system to assess the health status of wind turbines. Summary of the Invention

[0003] The purpose of this invention is to provide a method for assessing the health status of wind turbine generators that takes into account operating conditions. This method provides more objective and accurate assessment results and can detect the deterioration trend of wind turbine generators in advance.

[0004] The technical solution adopted in this invention is a method for assessing the health status of wind turbine generators considering operating conditions, which is implemented according to the following steps:

[0005] Step 1: Preprocess the historical SCADA data of the wind turbine;

[0006] Step 2: Based on the preprocessed data, the mRMR algorithm is used to select the relevant feature parameters of each prediction parameter;

[0007] Step 3: Divide the operating conditions of the wind turbine units;

[0008] Step 4: Construct a multi-parameter prediction model based on CNN-BiGRU under various operating conditions;

[0009] Step 5: Divide the historical data of the wind turbine under healthy operation status into working conditions according to Step 3, divide the feature parameter data of each working condition into training set and test set, input them into CNN-BiGRU prediction model, obtain the predicted value, and then construct a standard residual set based on the residual between the predicted value and the actual value of the training set.

[0010] Step 6: Determine the health status level indicators;

[0011] Step 7: Determine the current health level of the wind turbine based on the real-time SCADA data.

[0012] The invention is further characterized by:

[0013] The data preprocessing process in step 1 is as follows:

[0014] First, data below the cut-off wind speed and those with negative power values ​​are removed. Then, the wind speed-power and power-wind speed data sequences are arranged in ascending order and divided into four equal parts using the quartile method. Each part represents 25% of the total data. The data at the 25%, 50%, and 75% thresholds are designated as quartiles, denoted as the first quartile (Q1), second quartile (Q2), and third quartile (Q3), respectively. The difference between the third quartile and the first quartile is the interquartile range (IQR).

[0015]

[0016] Will be in ~ Data outside the specified range is removed.

[0017] In step 2, power, gearbox bearing I temperature, and generator bearing temperature are used as multi-parameter prediction targets. The mRMR algorithm is used to screen the candidate parameters, and the candidate parameters with the top six mRMR values ​​are selected as the relevant feature parameters of each prediction parameter.

[0018] Step 3 specifically involves:

[0019] Using wind speed, generator speed, and torque as the basis for classifying operating conditions, the fuzzy C-means clustering algorithm is used to cluster the data after preprocessing in step 1, thereby classifying the operating conditions of the wind turbine and obtaining the cluster center for each operating condition.

[0020] Step 4 is as follows:

[0021] A multi-parameter prediction model based on CNN-BiGRU was constructed, and the model was trained based on the sample data of each working condition divided in step 3 to obtain the multi-parameter prediction model under each working condition. At the same time, the sparrow search algorithm was used to optimize the number of hidden nodes, the initial learning rate and the L2 regularization coefficient of the BiGRU network.

[0022] In step 4, the CNN-BiGRU prediction model includes an input layer, a CNN layer, a BiGRU layer, and an output layer. The kernel size is set to 3, the stride is 1, the activation function is ReLU, the pooling layer uses average pooling, the maximum number of training iterations is set to 100, and the Adam optimization algorithm is used.

[0023] In step 6, the health status level index is determined by the distribution of Mahalanobis distance between the residuals predicted by the multi-parameter prediction of the test set in step 5 and the standard residual set.

[0024] Step 6 specifically involves: calculating the Mahalanobis distance between the residuals of the predicted values ​​of multiple parameters in the test set under healthy operating conditions of the wind turbine and the standard residual set; performing a moving average and dynamic weighting on the obtained Mahalanobis distance; and defining the wind turbine health index based on the dynamically weighted Mahalanobis distance.

[0025]

[0026] The operating status of wind turbine units is divided into three levels: healthy, qualified, and abnormal, and health indicators are set. The value is the distribution of the Mahalanobis distance between the residuals of the multi-parameter prediction results under healthy operating conditions of the wind turbine and the standard residual set. The value is determined, and according to the 3σ criterion, the Mahalanobis distance at μ+3σ of the distribution is used as the dividing point between the healthy and qualified levels, and the maximum value of the distribution is used to determine the abnormal state index of the wind turbine.

[0027] Step 7 specifically includes:

[0028] The operating conditions of the current wind turbine are identified. The real-time SCADA data of the current wind turbine is input into the CNN-BiGRU prediction model under the corresponding operating conditions to obtain the predicted values. Then, the health index of the wind turbine is calculated based on the Mahalanobis distance between the residual of the real-time multi-parameter predicted values ​​and the actual values ​​of the wind turbine and the standard residual set. Finally, the current health level of the wind turbine is determined based on the results of step 6.

[0029] The beneficial effects of this invention are:

[0030] This invention eliminates the need for mathematical models of each subsystem of the wind turbine. It utilizes readily available SCADA data from the wind turbine, considers the operating conditions of the turbine, and employs a fuzzy C-means clustering algorithm to classify the operating conditions. Under various conditions, a multi-parameter CNN-BiGRU prediction model is constructed based on a bidirectional gated recurrent unit neural network. The CNN-BiGRU network parameters are optimized using a sparrow search algorithm. A standard residual set is constructed using the residuals between the predicted and actual values ​​of the multi-parameters under healthy operating conditions. A health assessment index is built based on the Mahalanobis distance between the residuals between the real-time predicted and actual values ​​of the multi-parameters and the standard residual set. This enables the assessment of the wind turbine's health status, resulting in more objective and accurate assessments that can detect deterioration trends in advance and provide early warnings. Attached Figure Description

[0031] Figure 1 This is a flowchart of the method of the present invention;

[0032] Figure 2 This is a diagram of the CNN-BiGRU network structure of this invention;

[0033] Figure 3 This is a scatter plot of wind speed and power before data preprocessing in Embodiment 1 of the present invention;

[0034] Figure 4 This is a scatter plot of wind speed and power after data preprocessing in Embodiment 1 of the present invention;

[0035] Figure 5 This is a diagram showing the result of the working condition division in Embodiment 1 of the present invention;

[0036] Figure 6 This is the curve showing the predicted result and actual monitored value of the power parameter under operating condition 1 in Embodiment 2 of the present invention;

[0037] Figure 7 This is the curve showing the predicted results and actual monitored values ​​of the power parameters under operating condition 2 in Embodiment 2 of the present invention;

[0038] Figure 8 This is the curve showing the predicted results and actual monitored values ​​of the three power parameters under operating conditions in Embodiment 2 of the present invention;

[0039] Figure 9 This is the curve showing the predicted results and actual monitored values ​​of the four power parameters under operating conditions in Embodiment 2 of the present invention;

[0040] Figure 10 It is the Mahalanobis distance between the real-time residual and the standard residual set in Embodiment 2 of the present invention;

[0041] Figure 11 This refers to the unit health indicators in Embodiment 2 of the present invention. Detailed Implementation

[0042] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0043] This invention considers a health status assessment method for wind turbine generators under operating conditions, such as... Figure 1 As shown, please follow these steps:

[0044] Step 1: Preprocess the data stored in the Supervisory Control and Data Acquisition (SCADA) system for wind turbines.

[0045] First, data with wind speeds below the cut-in speed and negative power values ​​are removed. Then, the quartile method is used to process the remaining historical data to remove outliers. The specific process is as follows:

[0046] The wind speed-power (VP) and power-wind speed (PV) data series were arranged in ascending order and divided into four equal parts using the quartile method. Each part accounted for 25% of the total data. The data points at the 25%, 50%, and 75% thresholds were called quartiles, denoted as the first quartile (Q1), second quartile (Q2), and third quartile (Q3), respectively. The difference between the third quartile and the first quartile is called the interquartile range (IQR).

[0047] (1)

[0048] In statistics, it is considered to be in ~ Data outside the specified range are outliers, so these outliers are removed to perform data preprocessing.

[0049] Step 2: Select 16 parameters, including wind speed, wind turbine speed, generator torque, and torque, from the preprocessed SCADA data of the wind turbine in Step 1. With power, gearbox bearing I temperature, and generator bearing temperature as the multi-parameter prediction targets, the Max-Relevance and Min-Redundancy (mRMR) algorithm is used to filter the 16 candidate parameters. The candidate parameters with the top six mRMR values ​​are selected as the relevant feature parameters of each prediction parameter.

[0050] The process for calculating the mRMR value is as follows:

[0051] First, calculate the mutual information between each candidate parameter and each predicted parameter according to equation (2):

[0052] (2)

[0053] In equation (2), , , Random variables , Marginal probability density and joint probability density.

[0054] Then, the maximum correlation index between the feature subset and each prediction parameter is calculated:

[0055] (3)

[0056] In equation (3), For feature subset, For the target variable, For the first One characteristic, The number of features.

[0057] Calculate the minimum redundancy index to minimize the correlation between the feature variables in the selected feature set:

[0058] (4)

[0059] Finally, taking into account both the maximum correlation index and the minimum redundancy index criteria, the operator is defined. The mRMR value is calculated as follows:

[0060] (5)

[0061] Step 3: Using wind speed, generator speed, and torque as the basis for dividing operating conditions, the Fuzzy C-means (FCM) clustering algorithm is used to cluster the data after preprocessing in Step 1, thereby dividing the operating conditions of the wind turbine and obtaining the cluster center for each operating condition.

[0062] FCM achieves soft clustering by introducing membership degrees, assuming a dataset matrix. X for:

[0063] (6)

[0064] In formula (6): n The total number of samples contained in the sample dataset. x i For the first i The feature data vector of each sample.

[0065] The objective function and constraints of FCM are as follows:

[0066] (7)

[0067] (8)

[0068] In equations (7) to (8): V As cluster center, U The membership matrix, c Indicates the number of cluster categories. n Indicates the number of samples. u ik For the first i The first sample k Class membership degree m For ambiguity, For the first i The sample to the first k Distance to cluster centers.

[0069] The general steps of FCM are as follows:

[0070] 1) Initialize parameters, given the number of clusters. c Ambiguity m Tolerance e and the number of iterations to terminate;

[0071] 2) Initialize the membership matrix U ;

[0072] 3) According to U j Calculate cluster centers V j ;

[0073] 4) Calculate the objective function J ;

[0074] 5) According to V j renew U j+1 (Return to step 3), and repeat until the end.

[0075] Step 4: Construct a multi-parameter prediction model based on CNN-BiGRU, and train the model based on the sample data of each working condition divided in Step 3 to obtain the multi-parameter prediction model under each working condition. At the same time, the Sparrow Search Algorithm (SSA) is used to optimize the number of hidden nodes, the initial learning rate and the L2 regularization coefficient of the BiGRU network.

[0076] The CNN-BiGRU prediction model is a network model that combines a Convolutional Neural Network (CNN) and a Bidirectional Gated Recurrent Unit (BiGRU). Its structure is as follows: Figure 2 As shown, the model includes an input layer, a CNN layer, a BiGRU layer, and an output layer. The prediction model is configured with a kernel size of 3, a stride of 1, a ReLU activation function, average pooling, a maximum training iteration count of 100, and uses the Adam optimization algorithm.

[0077] Step 5: Divide the historical data of the wind turbine under healthy operating conditions into operating conditions according to Step 3. Based on the feature parameters selected in Step 2, divide the feature parameter data under each operating condition into training and test sets and input them into the CNN-BiGRU prediction model trained in Step 4 to obtain predicted values. Then, construct a standard residual set based on the residuals between the predicted values ​​and actual values ​​in the training set, as shown in the following formula:

[0078] (9)

[0079] In equation (9), For the standard residual set, For predicted values, This is the actual value.

[0080] Step 6: Determine the health status level index based on the distribution of the Mahalanobis distance (MD) between the residuals predicted by the multi-parameter prediction of the test set in Step 5 and the standard residual set.

[0081] The Mahalanobis distance between the residuals of the predicted values ​​of the test set parameters and the standard residual set is calculated according to equation (10):

[0082] (10)

[0083] In equation (10), for Residuals from multi-parameter time-of-flight prediction For the standard residual set, and Distributions The mean and covariance.

[0084] To eliminate the influence of random errors, the size of the sliding window is set to m, and the Mahalanobis distance of the moving average at time t is obtained as follows:

[0085] (11)

[0086] The Mahalanobis distance of the multi-parameter prediction results after moving average is dynamically weighted, and the weight update formula is as follows:

[0087] (12)

[0088] Based on the Mahalanobis distance, the health index of wind turbines is defined as follows:

[0089] (13)

[0090] The operating status of wind turbine units is divided into three levels: healthy, qualified, and abnormal, and health indicators are set. The health level is determined by the distribution of Mahalanobis distances between the residuals of multi-parameter predictions and the standard residual set under healthy operating conditions of the wind turbine. The value is determined. According to the 3σ criterion, the Mahalanobis distance at μ+3σ of this distribution is used as the dividing point between the healthy and qualified levels, and the maximum value of this distribution is used to determine the abnormal state index of the wind turbine.

[0091] Step 7: Identify the current operating conditions of the wind turbine. Input the real-time SCADA data of the wind turbine into the CNN-BiGRU prediction model under the corresponding operating conditions to obtain the predicted values. Then, calculate the health index of the wind turbine based on the Mahalanobis distance between the residuals of the real-time multi-parameter predicted values ​​and the actual values ​​of the wind turbine and the standard residual set. Finally, determine the current health level of the wind turbine based on the results of Step 6.

[0092] Example 1:

[0093] This embodiment takes a 2MW wind turbine as an example and determines the health indicators based on its historical SCADA data:

[0094] The wind speed-power scatter plot of the preprocessed data obtained in this embodiment is shown below. Figure 3 As shown, the wind speed-power scatter plot of the preprocessed data is as follows: Figure 4 As shown, the results of the working condition classification using the fuzzy C-means clustering algorithm are as follows: Figure 5 As shown, the system is divided into four operating conditions. Tables 1-3 show the mRMR values ​​of different feature variables. The feature parameters with the top six mRMR values ​​were selected as the input features of the model.

[0095] Table 1. mRMR values ​​of power-related characteristic variables

[0096]

[0097] Table 2. mRMR values ​​of generator bearing temperature-related characteristic variables

[0098]

[0099] Table 3. Temperature-related characteristic variable mRMR values ​​of gearbox bearing I

[0100]

[0101] The results of optimizing the BiGRU network using the sparrow search algorithm are shown in Table 4.

[0102] Table 4 Optimization Results

[0103]

[0104] The operating status of wind turbine units is divided into three levels: healthy, qualified, and abnormal, and health indicators are set. The health level is determined by the distribution of Mahalanobis distances between the residuals of multi-parameter predictions and the standard residual set under healthy operating conditions of the wind turbine. The value was determined. Based on the 3σ criterion, the Mahalanobis distance at μ+3σ of this distribution was used as the dividing point between healthy and acceptable levels, resulting in α=12.74. The maximum value of this distribution was used to determine the index for abnormal state alarms of wind turbine units. The obtained health status indexes of the wind turbine units are shown in Table 5.

[0105] Table 5 Health Status Level Indicators of Wind Turbine Units

[0106]

[0107] Example 2:

[0108] This embodiment selects 5000 sets of actual SCADA monitoring data from the wind turbine unit in Example 1 under healthy operating conditions as training samples for training the multi-parameter prediction network. For example... Figures 6-9 The figures show the power parameter prediction results and actual monitoring value curves of the CNN-BiGRU network of the present invention under various operating conditions. It can be seen that the CNN-BiGRU network constructed in this invention can accurately predict the parameters of the wind turbine under various operating conditions.

[0109] The model was validated using historical data from the three days prior to the abnormal shutdown of the unit at 7:30 AM on June 8, 2017. Using the feature parameters selected according to the mRMR algorithm as inputs to the model, predictions were made for multiple feature parameters. The Mahalanobis distance between the real-time residuals and the standard residual set, and the corresponding health indicator values ​​were calculated as follows: Figure 10 , Figure 11 As shown, it can be seen that the Mahalanobis distance began to increase at 6:10 on June 7th. Figure 11 The wind turbine health index, determined by Mahalanobis distance, shows that starting from 1:40 AM on June 7th, the health index repeatedly dropped to the range of [0.86, 0.9]. At 12:10 PM on June 7th, the health index fell below 0.86, indicating an abnormal state of the wind turbine, until the turbine issued a shutdown alarm. The results show that the method of this invention detected the abnormal state of the wind turbine 19 hours earlier than the SCADA system.

[0110] Example 3:

[0111] The health status assessment method for wind turbine generators, taking into account operating conditions, is implemented according to the following steps:

[0112] Step 1: Preprocess the historical SCADA data of the wind turbine;

[0113] Step 2: Based on the preprocessed data, the mRMR algorithm is used to select the relevant feature parameters of each prediction parameter;

[0114] Step 3: Divide the operating conditions of the wind turbine units;

[0115] Step 4: Construct a multi-parameter prediction model based on CNN-BiGRU under various operating conditions;

[0116] Step 5: Divide the historical data of the wind turbine under healthy operation status into working conditions according to Step 3, divide the feature parameter data of each working condition into training set and test set, input them into CNN-BiGRU prediction model, obtain the predicted value, and then construct a standard residual set based on the residual between the predicted value and the actual value of the training set.

[0117] Step 6: Determine the health status level indicators;

[0118] Step 7: Determine the current health level of the wind turbine based on the real-time SCADA data.

[0119] Example 4:

[0120] Based on Example 3, the data preprocessing process in step 1 is as follows:

[0121] First, data below the cut-off wind speed and those with negative power values ​​are removed. Then, the wind speed-power and power-wind speed data sequences are arranged in ascending order and divided into four equal parts using the quartile method. Each part represents 25% of the total data. The data at the 25%, 50%, and 75% thresholds are designated as quartiles, denoted as the first quartile (Q1), second quartile (Q2), and third quartile (Q3), respectively. The difference between the third quartile and the first quartile is the interquartile range (IQR).

[0122]

[0123] Will be in ~ Data outside the specified range is removed.

[0124] Example 5:

[0125] Based on Example 4, in step 2, power, gearbox bearing I temperature, and generator bearing temperature are used as multi-parameter prediction targets. The mRMR algorithm is used to screen the candidate parameters, and the candidate parameters with the top six mRMR values ​​are selected as the relevant feature parameters of each prediction parameter.

[0126] Example 6:

[0127] Based on Example 5, step 3 specifically includes:

[0128] Using wind speed, generator speed, and torque as the basis for classifying operating conditions, the fuzzy C-means clustering algorithm is used to cluster the data after preprocessing in step 1, thereby classifying the operating conditions of the wind turbine and obtaining the cluster center for each operating condition.

Claims

1. A method for assessing the health status of wind turbine generators considering operating conditions, characterized in that, The specific steps are as follows: Step 1: Preprocess the historical SCADA data of the wind turbine; Step 2: Based on the preprocessed data, the mRMR algorithm is used to select the relevant feature parameters of each prediction parameter; Step 3: Divide the operating conditions of the wind turbine units; Step 4: Construct a multi-parameter prediction model based on CNN-BiGRU under various operating conditions; Step 5: Divide the historical data of the wind turbine under healthy operation status into working conditions according to Step 3, divide the feature parameter data of each working condition into training set and test set, input them into CNN-BiGRU prediction model, obtain the predicted value, and then construct a standard residual set based on the residual between the predicted value and the actual value of the training set. Step 6: Determine the health status level indicators; Step 7: Determine the current health level of the wind turbine based on the real-time SCADA data. Specifically, step 6 involves: calculating the Mahalanobis distance between the residuals of the predicted values ​​of multiple parameters in the test set under healthy operating conditions of the wind turbine and the standard residual set; performing a moving average and dynamic weighting on the obtained Mahalanobis distance; and defining the wind turbine health index based on the dynamically weighted Mahalanobis distance. ， The operating status of wind turbine units is divided into three levels: healthy, qualified, and abnormal, and health indicators are set. The value is the distribution of the Mahalanobis distance between the residuals of the multi-parameter prediction results under healthy operating conditions of the wind turbine and the standard residual set. The value is determined, and according to the 3σ criterion, the Mahalanobis distance at μ+3σ of the distribution is used as the dividing point between the healthy and qualified levels, and the maximum value of the distribution is used to determine the abnormal state index of the wind turbine.

2. The method for assessing the health status of wind turbine generators considering operating conditions according to claim 1, characterized in that, The data preprocessing process in step 1 is as follows: First, data below the cut-off wind speed and those with negative power values ​​are removed. Then, the wind speed-power and power-wind speed data sequences are arranged in ascending order and divided into four equal parts using the quartile method. Each part represents 25% of the total data. The data at the 25%, 50%, and 75% thresholds are designated as quartiles, denoted as the first quartile (Q1), second quartile (Q2), and third quartile (Q3), respectively. The difference between the third quartile and the first quartile is the interquartile range (IQR). ， Will be in ~ Data outside the specified range is removed.

3. The method for assessing the health status of wind turbine generators considering operating conditions according to claim 1, characterized in that, In step 2, power, gearbox bearing I temperature, and generator bearing temperature are used as multi-parameter prediction targets. The mRMR algorithm is used to screen the candidate parameters, and the candidate parameters with the top six mRMR values ​​are selected as the relevant feature parameters of each prediction parameter.

4. The method for assessing the health status of wind turbine generators considering operating conditions according to claim 1, characterized in that, Step 3 specifically involves: Using wind speed, generator speed, and torque as the basis for classifying operating conditions, the fuzzy C-means clustering algorithm is used to cluster the data after preprocessing in step 1, thereby classifying the operating conditions of the wind turbine and obtaining the cluster center for each operating condition.

5. The method for assessing the health status of wind turbine generators considering operating conditions according to claim 1, characterized in that, Step 4 is as follows: A multi-parameter prediction model based on CNN-BiGRU was constructed, and the model was trained based on the sample data of each working condition divided in step 3 to obtain the multi-parameter prediction model under each working condition. At the same time, the sparrow search algorithm was used to optimize the number of hidden nodes, the initial learning rate and the L2 regularization coefficient of the BiGRU network.

6. The method for assessing the health status of wind turbine generators considering operating conditions according to claim 1, characterized in that, In step 4, the CNN-BiGRU prediction model includes an input layer, a CNN layer, a BiGRU layer, and an output layer. The kernel size is set to 3, the stride is 1, the activation function is ReLU, the pooling layer uses average pooling, the maximum number of training iterations is set to 100, and the Adam optimization algorithm is used.

7. The method for assessing the health status of wind turbine generators considering operating conditions according to claim 1, characterized in that, Step 7 specifically includes: The operating conditions of the current wind turbine are identified. The real-time SCADA data of the current wind turbine is input into the CNN-BiGRU prediction model under the corresponding operating conditions to obtain the predicted values. Then, the health index of the wind turbine is calculated based on the Mahalanobis distance between the residual of the real-time multi-parameter predicted values ​​and the actual values ​​of the wind turbine and the standard residual set. Finally, the current health level of the wind turbine is determined based on the results of step 6.

Images