Wind turbine reliability assessment method and device based on dynamic LSTM-Sage-Husa algorithm

Abstract

The invention belongs to the field of wind turbine reliability assessment, and provides a wind turbine reliability assessment method and device based on a dynamic LSTM-Sage-Husa algorithm. The invention obtains operation data of various components of the wind turbine from a SCADA system, predicts performance parameters of a degradation process by using a multi-layer perception LSTM model, and adopts a dynamic Sage-Husa algorithm to correct a nonlinear Wiener degradation process, introduces a mean and a covariance matrix of a state vector, establishes a nonlinear Wiener degradation process of the wind turbine considering effective impact, and assesses the reliability of the wind turbine. The invention considers the influence of effective impact on the wind turbine, and by adopting a multi-layer perception LSTM model to predict the problem of lack of data in the degradation process, the problem of inaccurate reliability assessment of the wind turbine caused by data loss can be solved. In addition, in order to more accurately capture the performance change of the unit in the nonlinear degradation process and reduce the volatility of the data, the noise in the degradation data is removed, the distribution parameter variable is introduced, the statistical model and the distribution parameter in the degradation process are obtained by using the dynamic Sage-Husa algorithm, and the optimal estimated degradation function is obtained by combining the predicted distribution parameter, thereby improving the accuracy of the model.

Description

Technical Field

[0001] The present invention belongs to the field of wind turbine reliability assessment, and in particular relates to a wind turbine reliability assessment method and device based on a dynamic LSTM-Sage-Husa algorithm. Background Art

[0002] As the global demand for renewable energy continues to increase, wind energy, as a clean and renewable energy form, has occupied an increasingly important position in the energy structure of various countries. However, environmental changes and the aging of operating components have made the study of the degradation process of wind turbines a key issue. The degradation process of wind turbines is often nonlinear, and the performance parameters are also complex. Traditional methods cannot achieve the expected accuracy, and traditional methods require a large amount of fault data to describe the degradation process, which consumes a lot of costs. Therefore, how to accurately evaluate the reliability of wind turbines with a small amount of data has become an important issue. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides a wind turbine reliability assessment method based on a dynamic LSTM-Sage-Husa algorithm to solve the problems in the prior art. The technical solution adopted by the present invention is:

[0004] The wind turbine reliability assessment method based on the dynamic LSTM-Sage-Husa algorithm includes the following steps:

[0005] Step 1: Obtain the operation data of the wind turbine according to the SCADA system;

[0006] Step 2: Use the multi-layer perception LSTM model to predict performance parameters during the degradation process, including:

[0007] Different operating parameters obtained by the SCADA system in step 1 and the measurement parameter vector ;

[0008] Wherein, N is the number of sampling points during the operation of the wind turbine; W is the parameter vector of different working conditions; C is the measurement parameter vector; is the m-dimensional operating condition parameter vector, is the m-dimensional measurement parameter vector;

[0009] The length of the prediction sequence is set to n. By using an autoregressive model to establish a parameter prediction model for each set of one-dimensional measurement parameter features, the prediction process of the i-th regression model at time t is obtained:

[0010] ;

[0011] in: Predicted values ​​of measurement parameters obtained for historical data input; is the operating condition parameter vector; t is the time series; n is the length of the prediction sequence;

[0012] The multi-layer perception mechanism is used to calculate the attention score of each feature. Secondly, the Softmax function is used to convert the attention score into the attention weight. Finally, the attention weight is weighted on the input variable in the feature dimension to obtain the weighted input variable at time t:

[0013] ;

[0014] in: is the working condition parameter after the hidden state is updated. is the attention weight of the mth variable, is the m-th dimension operating condition parameter vector before updating, and m is the total number of variables;

[0015] Get the encoded feature sequence , and then the historical performance parameter sequence The results of the encoder output and the original working condition parameter sequence are spliced ​​in the feature dimension and imported into the LSTM model. The LSTM model learns the time connection of the sequence to complete the prediction of the performance parameters;

[0016] Step 3: Use dynamic Sage-Husa adaptive Kalman filter to fuse and correct wind turbine performance parameters;

[0017] Step 4: Establish a wind turbine reliability assessment model based on the improved nonlinear Wiener degradation process;

[0018] Step 5: Design a wind turbine reliability assessment device.

[0019] Furthermore, step three includes: obtaining the mean and covariance matrix of the approximate original state distribution by selecting sampling points in the original degraded state distribution:

[0020] ;

[0021] in: is the mean of the approximate original state distribution, is the covariance of the approximate original state distribution, is the mean function;

[0022] According to the state vector x, construct the sampling point , and determine the mean weight and covariance weight of each sampling point:

[0023] ;

[0024] in: Proportional correction factor, is the weight coefficient, is a constant weight parameter greater than zero, is the mean weight of the i-th sampling point, is the covariance weight of the i-th sampling point;

[0025] Update state variables, construct a sampling point set based on time t-1, and predict the sampling point at time t , and obtain the one-step prediction mean of the system state variables at time t and the covariance matrix :

[0026] ;

[0027] in: is the predicted mean of the system state variables at time t, is the covariance matrix of the system state variables at time t, is the system error of the state equation of the wind turbine, for The covariance matrix of is the sampling point at time t predicted based on the sampling point at time t-1;

[0028] According to statistics and Get the sampling point set , combined with the observation equation to obtain the observed quantity , and finally get the predicted mean of the observed variable at time t:

[0029] ;

[0030] in: is the predicted mean of the predictor variable, is the mean weight of the i-th sampling point;

[0031] Through the dynamic Sage-Husa adaptive unscented Kalman filtering method, the genetic factors are adjusted to obtain:

[0032] ;

[0033] in: is the observation residual vector matrix, is the coefficient of variation of genetic factors, is the measurement matrix, is the covariance matrix of the system state variables at time t, is the mean value of the degradation at time t, is the genetic factor that changes with time t;

[0034] Calculate the autocovariance and cross-covariance of the observed variables at time t:

[0035] ;

[0036] in: is the autocovariance of the predictor variables, The cross-covariance of the predictor variables, is the covariance weight of the i-th sampling point, for The covariance matrix of for The covariance matrix of

[0037] Integrate to obtain the optimal state estimate and covariance matrix:

[0038] ;

[0039] in: The optimal state estimate, is the optimal covariance matrix, is the predicted mean of the system state variables at time t, is the mean of the observed variable at time t, is the predicted mean of the predictor variable, is the Kalman gain, and ;

[0040] Combined with step 2, the state equation at time t is expressed as:

[0041] ;

[0042] in: is the system error of the state equation of the wind turbine, is the time scale function representing the nonlinearity of the degradation process, is the state equation of the degradation process at time t-1.

[0043] Furthermore, step four includes:

[0044] Modeling the degradation process:

[0045] ;

[0046] in: is the performance degradation of the wind turbine at time t, and when t=0, ; is the drift coefficient representing the degradation rate, is the diffusion coefficient, is the standard Brownian motion, is the time scale function representing the nonlinearity of the degradation process;

[0047] The composite Poisson process is used to represent the step degradation increment caused by the effective shock at time t on the degradation process. :

[0048] ;

[0049] in: is the step degradation increment caused by the effective impact on the degradation process at time t, is a composite Poisson process; is the effective impact amplitude, In order to cause the lowest effective impact amplitude of wind turbines, is the proportionality coefficient;

[0050] Introduce the degradation rate influencing factor:

[0051] ;

[0052] in: is the performance degradation rate of the wind turbine after being impacted, is the normal performance degradation rate of the wind turbine, To introduce the degradation rate influencing factor, is the impact correlation coefficient, is a compound Poisson distribution process;

[0053] The cumulative step degradation of the wind turbine and degradation rate influencing factors After combining, the degradation model considering the effective impact magnitude is obtained:

[0054] ;

[0055] When a wind turbine fails due to degradation, the life function is:

[0056] ;

[0057] in: is the failure threshold, is the infimum function, is the life function of the wind turbine;

[0058] The reliability function of the wind turbine is obtained by integration:

[0059] ;

[0060] in: is the reliability function of the wind turbine, is a probability function.

[0061] Furthermore, step five includes: integrating steps one to four into a wind turbine reliability assessment device.

[0062] A wind turbine reliability assessment device based on a dynamic LSTM-Sage-Husa algorithm, the device comprising: a data collection module, a performance parameter prediction module and a degradation model correction module;

[0063] Data preprocessing module: communicates with the SCADA system to obtain the operating data and environmental data of each component of the wind turbine, and obtains the performance degradation data of the wind turbine by summarizing and sorting;

[0064] Missing data prediction module: import the degradation data into the multi-layer perception LSTM model to predict the missing data, and apply the missing data obtained in the nonlinear Wiener degradation process;

[0065] Degradation model correction module: The dynamic Sage-Husa adaptive unscented Kalman filter model is used to select the optimal parameters in the degradation process, and a nonlinear Wiener degradation process model is established to evaluate its reliability.

[0066] The present invention has the following beneficial effects: the present invention establishes a nonlinear Wiener degradation process for the life of a wind turbine generator set, takes into account the influence of effective impact on the wind turbine generator set, and adopts a multi-layer perception LSTM model to predict the problem of lack of data in the degradation process, thereby effectively solving the problem of inaccurate reliability assessment of the wind turbine generator set due to missing data. Then, a distribution parameter variable is introduced, and a dynamic Sage-Husa algorithm is used to obtain a statistical model and distribution parameters in the degradation process, and combined with the predicted distribution parameters, an optimal estimated degradation function is obtained, which more accurately captures the performance changes of the unit in the nonlinear degradation process, reduces the volatility of the data, removes the noise in the degradation data, and improves the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the embodiments of the present invention Figure 1 , the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0069] The invention provides a wind turbine reliability assessment method and device based on a dynamic LSTM-Sage-Husa algorithm. The invention first obtains the operation data of each component of the wind turbine from a SCADA system, then uses a multi-layer perception LSTM model to predict the degradation process performance parameters, adopts a dynamic Sage-Husa algorithm to correct the nonlinear Wiener degradation process, introduces the mean and covariance matrix of the state vector, then establishes the nonlinear Wiener degradation process of the wind turbine considering the effective impact, assesses the reliability of the wind turbine, and finally designs a wind turbine reliability assessment device according to the above steps.

[0070] Step 1: Get the operation data of wind turbines according to the SCADA system:

[0071] The detection of wind turbine operating data is one of the core functions of the wind farm SCADA (Supervisory Control and Data Acquisition) system, which can monitor the operating status of wind turbines in real time, collect data and diagnose faults. The SCADA system collects various operating parameters of wind turbines through sensors and data acquisition devices. Common detection indicators include: wind speed and direction, speed, power output, temperature, vibration, current and voltage. Then, the collected wind turbine operating data is transmitted to the data center of the SCADA system through the field bus or wireless network. The SCADA system usually uses a database for data storage and ensures the reliability and real-time performance of data transmission.

[0072] By analyzing historical data on the wind turbine operating environment, output status, etc., the performance degradation of the wind turbine and the aging of components are identified, and a data basis is provided for step 2 to obtain the distribution parameters of each component during the degradation process.

[0073] Step 2: Use the multi-layer perception LSTM model to predict performance parameters during degradation:

[0074] Different operating parameters obtained by the SCADA system in step 1 and the measurement parameter vector , where N is the number of sampling points during the operation of the wind turbine; W is the parameter vector of different working conditions; C is the measurement parameter vector; is the m-dimensional operating condition parameter vector, is an m-dimensional measurement parameter vector.

[0075] The length of the prediction sequence is set to n. By using an autoregressive model to establish a parameter prediction model for each set of one-dimensional measurement parameter features, the prediction process of the i-th regression model at time t is obtained:

[0076] in: Predicted values ​​of measurement parameters obtained for historical data input; is the operating condition parameter vector; t is the time series; n is the length of the prediction sequence.

[0077] Based on the above model, combined with the feature attention mechanism, the performance parameters in the wind turbine degradation model are predicted. This improved network model contains an encoder, a feature concatenation layer, and a decoder. The encoder part only performs the above working condition parameters. Encode, and then learn the connection between the working condition parameter features through the feature attention mechanism.

[0078] First, the multi-layer perception mechanism is used to calculate the attention score of each feature. Second, the Softmax function is used to convert the attention score into an attention weight. Finally, the attention weight is weighted on the input variable in the feature dimension to obtain the weighted input variable at time t:

[0079] ;

[0080] in: is the working condition parameter after the hidden state is updated, is the attention weight of the mth variable, is the m-th dimension operating condition parameter vector before updating, and m is the total number of variables;

[0081] Get the encoded feature sequence , and then the historical performance parameter sequence The results of the encoder output and the original working condition parameter sequence are spliced ​​in the feature dimension and imported into the LSTM model. The LSTM model is used to learn the time connection of the sequence and predict the performance parameters.

[0082] Step 3: Use dynamic Sage-Husa adaptive Kalman filter to integrate and correct wind turbine performance parameters:

[0083] Since the traditional Kalman filter assumes that the state equation and observation equation are linear, it has limitations in dealing with nonlinear system problems. Therefore, the unscented Kalman filter is used to deal with nonlinear systems, and a set of sampling points is used to capture the non-Gaussian characteristics of the state distribution.

[0084] By selecting sampling points in the original degraded state distribution according to certain rules, we can obtain the mean and covariance matrix of the approximate original state distribution:

[0085] ;

[0086] in: is the mean of the approximate original state distribution, is the covariance of the approximate original state distribution, is the mean function.

[0087] According to the state vector x, construct the sampling point , and determine the mean weight and covariance weight of each sampling point.

[0088] ;

[0089] in: Proportional correction factor, is the weight coefficient, is a constant weight parameter greater than zero, is the mean weight of the i-th sampling point, is the covariance weight of the i-th sampling point.

[0090] Update state variables, construct a sampling point set based on time t-1, and predict the sampling point at time t , the one-step prediction mean of the system state variable at time t is obtained through the above formula and the covariance matrix :

[0091] ;

[0092] in: is the predicted mean of the system state variables at time t, is the covariance matrix of the system state variables at time t, is the system error of the state equation of the wind turbine, for The covariance matrix of is the sampling point at time t predicted based on the sampling point at time t-1.

[0093] According to statistics and Get the sampling point set , combined with the observation equation to obtain the observed quantity , and finally get the predicted mean of the observed variable at time t:

[0094] ;

[0095] in: is the predicted mean of the predictor variable, is the mean weight of the i-th sampling point.

[0096] Through the dynamic Sage-Husa adaptive unscented Kalman filtering method, the genetic factors are dynamically adjusted to obtain:

[0097] ;

[0098] in: is the observation residual vector matrix, is the coefficient of variation of genetic factors, is the measurement matrix, is the covariance matrix of the system state variables at time t, is the mean value of the degradation at time t, is the genetic factor that changes with time t.

[0099] Calculate the autocovariance and cross-covariance of the observed variables at time t:

[0100] ;

[0101] in: is the autocovariance of the predictor variables, The cross-covariance of the predictor variables, is the covariance weight of the i-th sampling point, for The covariance matrix of . for The covariance matrix of .

[0102] Based on the above, we get the optimal state estimate and covariance matrix:

[0103] ;

[0104] in: The optimal state estimate, is the optimal covariance matrix, is the predicted mean of the system state variables at time t, is the mean of the observed variable at time t, is the predicted mean of the predictor variable, is the Kalman gain, and .

[0105] Combined with step 2, the state equation at time t is expressed as:

[0106] ;

[0107] in: is the system error of the state equation of the wind turbine, is the time scale function representing the nonlinearity of the degradation process, is the state equation of the degradation process at time t-1.

[0108] Step 4: Establish a wind turbine reliability assessment model based on the improved nonlinear Wiener degradation process:

[0109] The degradation process is modeled using a nonlinear Wiener process:

[0110] ;

[0111] in: is the performance degradation of the wind turbine at time t, and when t=0, . is the drift coefficient representing the degradation rate, is the diffusion coefficient, is the standard Brownian motion, is the time scale function representing the nonlinearity of the degradation process.

[0112] When the performance degradation of the wind turbine First time exceeding critical threshold The time T is called the life of the system degradation failure, and after being subjected to an effective impact, the degradation amount and degradation rate of the wind turbine degradation process will change.

[0113] Different types of effective impacts cause different damages to the degradation process. The composite Poisson process is used to represent the step degradation increment caused by the effective impact on the degradation process at time t. :

[0114] ;

[0115] in: is the step degradation increment caused by the effective impact on the degradation process at time t, is a compound Poisson process. is the effective impact amplitude, In order to cause the lowest effective impact amplitude of wind turbines, is the proportionality coefficient;

[0116] The effective impact not only affects the degradation increment of the wind turbine, but also changes the degradation rate. By introducing the degradation rate influencing factor:

[0117] ;

[0118] in: is the performance degradation rate of the wind turbine after being impacted, is the normal performance degradation rate of the wind turbine, To introduce the degradation rate influencing factor, is the impact correlation coefficient, is a compound Poisson distribution process.

[0119] The cumulative step degradation of the wind turbine and degradation rate influencing factors After combining, the degradation model considering the effective impact magnitude is obtained:

[0120] ;

[0121] When a wind turbine fails due to degradation, the life function is:

[0122] ;

[0123] in: is the failure threshold, is the infimum function, is the life function of the wind turbine.

[0124] Based on the above, the reliability function of the wind turbine is obtained:

[0125] ;

[0126] in: is the reliability function of the wind turbine, is a probability function.

[0127] Step 5: Based on the above steps, design a wind turbine reliability assessment device:

[0128] The above steps are integrated into the wind turbine reliability assessment device.

[0129] The wind turbine reliability assessment device based on the dynamic LSTM-Sage-Husa algorithm consists of three modules: data collection module, performance parameter prediction module and degradation model correction module. The following are the working contents of each module:

[0130] Data preprocessing module: communicates with the SCADA system to obtain the operating data of each component of the wind turbine and the environmental data. By summarizing and collating the above data, the performance degradation data of the wind turbine is obtained.

[0131] Missing data prediction module: Import complex degradation data into the multi-layer perception LSTM model to predict missing data, and apply the missing data obtained in the nonlinear Wiener degradation process.

[0132] Degradation model correction module: The dynamic Sage-Husa adaptive unscented Kalman filter model is used to select the optimal parameters in the degradation process, and a nonlinear Wiener degradation process model is established to evaluate its reliability.

[0133] The embodiments described above are only descriptions of the preferred modes of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations, modifications, and substitutions made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A wind turbine reliability assessment method based on the dynamic LSTM-Sage-Husa algorithm is characterized by: The following steps are involved: Step 1: Obtain the operation data of the wind turbine according to the SCADA system; Step 2: Use the multi-layer perception LSTM model to predict performance parameters during the degradation process, including: Different operating parameters obtained by the SCADA system in step 1 and the measurement parameter vector ; Wherein, N is the number of sampling points during the operation of the wind turbine; W is the parameter vector of different working conditions; C is the measurement parameter vector; is the m-dimensional operating condition parameter vector, is the m-dimensional measurement parameter vector; The length of the prediction sequence is set to n. By using an autoregressive model to establish a parameter prediction model for each set of one-dimensional measurement parameter features, the prediction process of the i-th regression model at time t is obtained: ； in: Predicted values ​​of measurement parameters obtained for historical data input; is the operating condition parameter vector; t is the time series; n is the length of the prediction sequence; The multi-layer perception mechanism is used to calculate the attention score of each feature. Secondly, the Softmax function is used to convert the attention score into the attention weight. Finally, the attention weight is weighted on the input variable in the feature dimension to obtain the weighted input variable at time t: ； in: is the working condition parameter after the hidden state is updated. is the attention weight of the mth variable, is the m-th dimension operating condition parameter vector before updating, and m is the total number of variables; Get the encoded feature sequence , and then the historical performance parameter sequence The results of the encoder output and the original working condition parameter sequence are spliced ​​in the feature dimension and imported into the LSTM model. The LSTM model learns the time connection of the sequence to complete the prediction of the performance parameters; Step 3: Use dynamic Sage-Husa adaptive Kalman filter to fuse and correct wind turbine performance parameters; Step 4: Establish a wind turbine reliability assessment model based on the improved nonlinear Wiener degradation process; Step 5: Design a wind turbine reliability assessment device.

2. The wind turbine reliability assessment method based on the dynamic LSTM-Sage-Husa algorithm according to claim 1 is characterized in that: Step three includes: obtaining the mean and covariance matrix of the approximate original state distribution by selecting sampling points in the original degraded state distribution: ； in: is the mean of the approximate original state distribution, is the covariance of the approximate original state distribution, is the mean function; According to the state vector x, construct the sampling point , and determine the mean weight and covariance weight of each sampling point: ； in: Proportional correction factor, is the weight coefficient, is a constant weight parameter greater than zero, is the mean weight of the i-th sampling point, is the covariance weight of the i-th sampling point; Update state variables, construct a sampling point set based on time t-1, and predict the sampling point at time t , and obtain the one-step prediction mean of the system state variables at time t and the covariance matrix : ； in: is the predicted mean of the system state variables at time t, is the covariance matrix of the system state variables at time t, is the system error of the state equation of the wind turbine, for The covariance matrix of is the sampling point at time t predicted based on the sampling point at time t-1; According to statistics and Get the sampling point set , combined with the observation equation to obtain the observed quantity , and finally get the predicted mean of the observed variable at time t: ； in: is the predicted mean of the predictor variable, is the mean weight of the i-th sampling point; Through the dynamic Sage-Husa adaptive unscented Kalman filtering method, the genetic factors are adjusted to obtain: ； in: is the observation residual vector matrix, is the coefficient of variation of genetic factors, is the measurement matrix, is the covariance matrix of the system state variables at time t, is the mean value of the degradation at time t, is the genetic factor that changes with time t; Calculate the autocovariance and cross-covariance of the observed variables at time t: ； in: is the autocovariance of the predictor variables, The cross-covariance of the predictor variables, is the covariance weight of the i-th sampling point, for The covariance matrix of for The covariance matrix of Integrate to obtain the optimal state estimate and covariance matrix: ； in: The optimal state estimate, is the optimal covariance matrix, is the predicted mean of the system state variables at time t, is the mean of the observed variable at time t, is the predicted mean of the predictor variable, is the Kalman gain, and ; Combined with step 2, the state equation at time t is expressed as: ； in: is the system error of the state equation of the wind turbine, is the time scale function representing the nonlinearity of the degradation process, is the state equation of the degradation process at time t-1.

3. The wind turbine reliability assessment method based on the dynamic LSTM-Sage-Husa algorithm according to claim 1 is characterized in that: Step 4 includes: Modeling the degradation process: ； in: is the performance degradation of the wind turbine at time t, and when t=0, ; is the drift coefficient representing the degradation rate, is the diffusion coefficient, is the standard Brownian motion, is the time scale function representing the nonlinearity of the degradation process; The composite Poisson process is used to represent the step degradation increment caused by the effective shock at time t on the degradation process. : ； in: is the step degradation increment caused by the effective impact on the degradation process at time t, is a composite Poisson process; is the effective impact amplitude, In order to cause the lowest effective impact amplitude of wind turbines, is the proportionality coefficient; Introduce the degradation rate influencing factor: ； in: is the performance degradation rate of the wind turbine after being impacted, is the normal performance degradation rate of the wind turbine, To introduce the degradation rate influencing factor, is the impact correlation coefficient, is a compound Poisson distribution process; The cumulative step degradation of the wind turbine and degradation rate influencing factors After combining, the degradation model considering the effective impact magnitude is obtained: ； When a wind turbine fails due to degradation, the life function is: ； in: is the failure threshold, is the infimum function, is the life function of the wind turbine; The reliability function of the wind turbine is obtained by integration: ； in: is the reliability function of the wind turbine, is a probability function.

4. The wind turbine reliability assessment method based on the dynamic LSTM-Sage-Husa algorithm according to any one of claims 1 to 3, characterized in that: Step five includes: integrating steps one to four into a wind turbine reliability assessment device.

5. A wind turbine reliability assessment device based on a dynamic LSTM-Sage-Husa algorithm, adopting the wind turbine reliability assessment method based on a dynamic LSTM-Sage-Husa algorithm as claimed in claim 4, characterized in that: The device comprises: a data collection module, a performance parameter prediction module and a degradation model correction module; Data preprocessing module: communicates with the SCADA system to obtain the operating data and environmental data of each component of the wind turbine, and obtains the performance degradation data of the wind turbine by summarizing and sorting; Missing data prediction module: import the degradation data into the multi-layer perception LSTM model to predict the missing data, and apply the missing data obtained in the nonlinear Wiener degradation process; Degradation model correction module: The dynamic Sage-Husa adaptive unscented Kalman filter model is used to select the optimal parameters in the degradation process, and a nonlinear Wiener degradation process model is established to evaluate its reliability.

Images