Reliability Evaluation Method for Wind Turbine Generator Sets Based on Improved ILOF-LSTM-Gamma Algorithm

Abstract

The present invention discloses a method for reliability assessment of wind turbine units based on an improved ILOF-LSTM-Gamma algorithm, which relates to the field of power technology and includes: obtaining historical state data of key components of wind turbine units through a SCADA system; preprocessing the historical state data by using the ILOF algorithm to obtain preprocessed data; establishing a performance degradation prediction model of wind turbine units based on an improved adaptive multi-head attention mechanism and an improved LSTM network method, inputting the preprocessed data into the performance degradation prediction model of wind turbine units to obtain predicted performance data of wind turbine units; establishing a reliability assessment model of wind turbine units by using a double-layer improved non-linear Gamma process, and inputting the predicted performance data of wind turbine units into the reliability assessment model of wind turbine units to obtain a reliability assessment result of wind turbine units. The present invention improves the accuracy of performance degradation prediction of wind turbine units and the accuracy of the assessment result.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric power, and particularly relates to a method for evaluating the reliability of a wind turbine based on an improved ILOF-LSTM-Gamma algorithm. Background Art

[0002] With the transformation of the global energy structure, wind power generation, as a clean and renewable energy source, is gradually increasing its proportion in the power system. However, wind turbines operate in complex and variable environmental conditions for a long time, and their key components (such as gearboxes, generators, blades, etc.) are easily affected by mechanical fatigue, environmental erosion, and random load changes, resulting in performance degradation or even failure. Moreover, the existing methods for evaluating the reliability of wind turbines rely on the structural and material characteristics of the turbines. Due to the large amount of data, the calculation is complex and difficult to adapt to actual working conditions, and the ability to process abnormal data is limited and easily affected by noise interference. In addition, when the existing Gamma process model deals with the uncertainty of the performance degradation of wind turbines, it is difficult to select unknown parameters and accurately characterize complex non-linear degradation trends. Therefore, an efficient and accurate method for evaluating the reliability of wind turbines is needed. Summary of the Invention

[0003] To solve the above technical problems, the present invention proposes a method for evaluating the reliability of a wind turbine based on an improved ILOF-LSTM-Gamma algorithm, which can make full use of the data of the SCADA system, effectively process abnormal data, and combine advanced deep learning and probability statistical models to improve the accuracy of predicting the performance degradation and evaluating the reliability of wind turbines.

[0004] To achieve the above object, the present invention provides a method for evaluating the reliability of a wind turbine based on an improved ILOF-LSTM-Gamma algorithm, including:

[0005] Obtaining historical status data of each key component of the wind turbine through the SCADA system;

[0006] Preprocessing the historical status data by using an improved preprocessing local outlier factor algorithm to obtain preprocessed data;

[0007] Establishing a wind turbine performance degradation prediction model based on an improved adaptive multi-head attention mechanism and an improved LSTM network method, and inputting the preprocessed data into the wind turbine performance degradation prediction model to obtain predicted wind turbine performance data;

[0008] Establishing a wind turbine reliability evaluation model by using a double-layer improved non-linear Gamma process, and inputting the predicted wind turbine performance data into the wind turbine reliability evaluation model to obtain a wind turbine reliability evaluation result.

[0009] Technical effects of the present invention: The present invention discloses a reliability evaluation method for wind turbine units based on an improved ILOF-LSTM-Gamma algorithm. First, by using the improved isolation forest outlier factor (ILOF) algorithm, outliers in the SCADA system data can be effectively identified and removed, reducing noise interference, ensuring the accuracy of the input data, and improving the reliability of the subsequent prediction model. Secondly, by combining the improved adaptive multi-head attention mechanism with the improved long short-term memory network (LSTM), the temporal characteristics of the wind turbine unit operation data can be fully captured, improving the modeling ability for long-term dependence relationships and enhancing the prediction accuracy of the performance degradation trend of key components. Finally, by using a two-layer improved non-linear Gamma process, the complexity of the wind turbine unit performance degradation can be more accurately described, overcoming the limitation of the traditional Gamma process in dealing with non-linear degradation and improving the applicability and robustness of the reliability evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings, which form a part of this application, are used to provide a further understanding of this application. The schematic embodiments and descriptions thereof of this application are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0011] Figure 1 are the historical state data of each key component of the wind turbine unit in the embodiment of the present invention;

[0012] Figure 2 is a schematic flowchart of the reliability evaluation method for wind turbine units based on the improved ILOF-LSTM-Gamma algorithm in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0013] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the accompanying drawings and combine the embodiments to detail this application.

[0014] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0015] As Figure 1 - Figure 2 shown, in this embodiment, a reliability evaluation method for wind turbine units based on the improved ILOF-LSTM-Gamma algorithm is provided, including:

[0016] Obtain the historical state data of each key component of the wind turbine unit through the SCADA system;

[0017] Preprocess the historical state data by using the improved isolation forest outlier factor algorithm to obtain preprocessed data;

[0018] Based on an improved adaptive multi - head attention mechanism and an improved LSTM network method, a performance degradation prediction model of a wind turbine is established. The pre - processed data is input into the performance degradation prediction model of the wind turbine to obtain the predicted performance data of the wind turbine.

[0019] An improved double - layer non - linear Gamma process is used to establish a reliability evaluation model of the wind turbine. The predicted performance data of the wind turbine is input into the reliability evaluation model of the wind turbine to obtain the reliability evaluation result of the wind turbine.

[0020] Furthermore, each key component of the wind turbine includes a gearbox component, a generator component, a pitch system component, and a yaw system component.

[0021] The historical state data of the gearbox component includes oil temperature, vibration, and gear wear state.

[0022] The historical state data of the generator component includes voltage, current, temperature, and power output.

[0023] The historical state data of the pitch system component includes pitch angle, control signal, and motor current.

[0024] The historical state data of the yaw system component includes yaw angle and yaw motor working state.

[0025] Furthermore, an improved pre - processing local outlier factor algorithm is used to pre - process the historical state data. The pre - processed data obtained includes:

[0026] An improved cubic spline interpolation method is used to supplement the historical state data to obtain a supplementary data set.

[0027] Based on the supplementary data set, an outlier detection algorithm is used to remove the data that does not conform to physical meaning, and the improved local outlier factor of each data point is obtained. If the improved local outlier factor is greater than the preset outlier threshold, this data point is an outlier, and the pre - processed data is obtained.

[0028] Furthermore, using an improved cubic spline interpolation method to supplement the historical state data includes:

[0029] ;

[0030] The constraint conditions are:

[0031] ;

[0032] ;

[0033] ;

[0034] Among them, is a cubic spline interpolation function; , and are interpolation coefficients to be determined respectively; is the status data of different components, are different components; is the th interpolation position of the th component; is the th interpolation data of the th component; is the first derivative of the cubic spline interpolation function at the th interpolation position of the th component; is the second derivative of the cubic spline interpolation function at the th interpolation position of the th component; is the data length of each component.

[0035] Furthermore, based on the supplementary data set, an outlier detection algorithm is used to remove data that does not conform to physical meaning, and the improved local outlier factor of each data point includes:

[0036] The Euclidean distance between point S and point O in the n-dimensional space is:

[0037] ;

[0038] Among them, is the distance from data point to data point in the historical status data of the th component; is based on the different health status ranges of each component; is the point at a distance of from data point in the neighborhood of data point ;

[0039] Calculate the improved local reachability density as:

[0040] ;

[0041] Among them, is the improved local reachability density; is the th reachability distance from data point to data point in the neighborhood of data point ; is the neighborhood length of the data point ;

[0042] Calculate the improved local outlier factor for each component status data:

[0043] ;

[0044] Among them, is the improved local outlier factor of the data point ; is the improved local reachability density of the data point ; is the improved local reachability density of the data point ;

[0045] Furthermore, establishing a wind turbine performance degradation prediction model based on the improved adaptive multi-head attention mechanism and the improved LSTM network method includes:

[0046] Import the processed historical data into the improved convolutional neural network to predict the performance degradation process of the wind turbine. Based on the Transformer architecture, first, multiple parallel attention heads encode and decode the data to identify important features, improve the long-term dependence modeling ability, and obtain local correlations. Secondly, the improved LSTM method is used to predict the performance degradation trend of the wind turbine.

[0047] The processed data is linearly transformed to convert the extracted feature data into queries, keys, and values. The similarity between each group of performance degradation data of the wind turbine is measured by the dot product of the query vector and the key vector, and the softmax function normalizes the attention scores to obtain the attention weights:

[0048] ;

[0049] Among them, is the attention weight, is the key vector, is the dimension of the key vector, is the query vector, and B is the adaptive bias matrix to control the attention allocation.

[0050] The data features , the th head in the multi-head self-attention respectively maps the central node feature and the adjacent node feature to the query vector and the key vector through different trainable parameters , and then according to the query vector and the key vector The attention distribution vector of the th head

[0051] ;

[0052] ;

[0053] ;

[0054] Among them, is the dot product function, calculating the correlation between the query vector and the key vector; l is the dimension of each self-attention head, and

[0055] is the total number of components of the wind turbine. The output of each head is calculated as:

[0056] ;

[0057] Among them, is the output of each head; is the adaptive weight of each head; is the value vector of each head.

[0058] Concatenate the outputs of all heads to obtain the final wind turbine performance degradation feature data:

[0059] ;

[0060] Among them, is the wind turbine performance degradation data processed by the improved adaptive multi-head attention mechanism; is the connection function; is the learnable weight output matrix.

[0061] Import the processed performance degradation data into the forget gate, input gate, and output gate to predict the wind turbine performance degradation function.

[0062] Calculation of the forget gate:

[0063] ;

[0064] Calculation of the input gate:

[0065] ;

[0066] Calculation of the output gate:

[0067] ;

[0068] Among them, , and are the results of the forget gate, input gate, and output gate respectively, is the output at time t-1, , and are the weights of each gate respectively, which can be adaptively adjusted according to the complexity of each group of data, , and are all the biases of each gate.

[0069] It is transformed into a predicted value through a fully connected layer:

[0070] ;

[0071] wherein, is the predicted data of the wind turbine, and are the parameters of the output layer.

[0072] Furthermore, establishing a reliability evaluation model for wind turbines using a two-layer improved non-linear Gamma process includes:

[0073] Establishing an upper-layer wind turbine evaluation model based on the improved non-linear Gamma process;

[0074] Establishing a lower-layer optimization model for the unknown parameters of the non-linear Gamma process based on the improved raccoon optimization algorithm;

[0075] Furthermore, establishing an upper-layer wind turbine evaluation model based on the improved non-linear Gamma process includes:

[0076] Importing the wind turbine performance degradation data predicted in step 3 into the improved non-linear Gamma process:

[0077] ;

[0078] wherein, , are the shape parameters respectively; is a continuous non-linear function with respect to t; is a scale random function; is an adjustable parameter; is the scale parameter;

[0079] ;

[0080] wherein, is the Gamma function; is the probability density function of the performance degradation process of each component of the wind turbine;

[0081] Furthermore, the obtained distribution function is as follows:

[0082] ;

[0083] Set the overall failure threshold of the wind turbine to the time when each key component reaches the failure threshold, and obtain the evaluation model of the wind turbine reliability as:

[0084] ;

[0085] Furthermore, the lower-layer optimization model for the unknown parameters of the nonlinear Gamma process based on the improved raccoon optimization algorithm includes:

[0086] ;

[0087] where is a random number between 0 and 1; is the position function of the unknown parameters of the nonlinear Gamma process; takes values from the set of unknown parameters { }; is the standard normal distribution function; is the optimal position; is a random integer; M is the number of raccoons;

[0088] ;

[0089] The following method is used for position update:

[0090] ;

[0091] where , are respectively the lower and upper bounds of the parameters varying with the iteration number .

[0092] The present invention discloses a method for evaluating the reliability of a wind turbine based on an improved ILOF-LSTM-Gamma algorithm. First, the improved isolation forest outlier factor (ILOF) algorithm is adopted, which can effectively identify and remove outliers in the SCADA system data, reduce noise interference, ensure the accuracy of the input data, and improve the reliability of the subsequent prediction model; second, combining the improved adaptive multi-head attention mechanism with the improved long short-term memory network (LSTM) can fully capture the temporal characteristics of the wind turbine operation data, improve the modeling ability for long-term dependence relationships, and enhance the prediction accuracy of the performance degradation trend of key components. Finally, the double-layer improved nonlinear Gamma process is adopted, which can more accurately describe the complexity of the performance degradation of the wind turbine, overcome the limitation that the traditional Gamma process is difficult to handle nonlinear degradation, and improve the applicability and robustness of the reliability evaluation.

[0093] The above are only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A wind turbine reliability assessment method based on the improved ILOF-LSTM-Gamma algorithm is characterized by: include: Obtain historical status data of key components of wind turbines through the SCADA system; Preprocessing the historical state data using an improved pre-local anomaly factor algorithm to obtain preprocessed data; Establishing a wind turbine performance degradation prediction model based on an improved adaptive multi-head attention mechanism and an improved LSTM network method, inputting the preprocessed data into the wind turbine performance degradation prediction model, and obtaining predicted wind turbine performance data; A wind turbine reliability assessment model is established by adopting a double-layer improved nonlinear Gamma process, and the predicted wind turbine performance data is input into the wind turbine reliability assessment model to obtain a wind turbine reliability assessment result; The key components of the wind turbine generator set include gearbox components, generator components, pitch system components and yaw system components; The historical status data of the gearbox components include oil temperature, vibration, and gear wear status; The historical status data of the generator components include voltage, current, temperature, and power output; The historical status data of the pitch system components include pitch angle, control signal, and motor current; The historical status data of the yaw system components include the yaw angle and the working status of the yaw motor; The improved pre-local anomaly factor algorithm is used to pre-process the historical state data, and the pre-processed data obtained includes: Using an improved cubic spline interpolation method to supplement the historical state data to obtain a supplementary data set; Based on the supplementary data set, an outlier detection algorithm is used to remove data that does not conform to physical meaning, and an improved local outlier factor of each data point is obtained. If the improved local outlier factor is greater than a preset outlier threshold, the data point is an outlier, and preprocessed data is obtained; The improved cubic spline interpolation method is used to supplement the historical state data, including: ； The constraints are: ； ； ； in, is the cubic spline interpolation function; , and are the interpolation coefficients to be determined respectively; Status data for different components, For different parts; For the Part No. interpolation positions; For the Part No. interpolation data; is the cubic spline interpolation function in the Part No. The first derivative at the interpolation position; is the cubic spline interpolation function in the Part No. The second derivative at the interpolation position; The data length of each component; Based on the supplementary data set, an outlier detection algorithm is used to remove data that does not conform to physical meaning, and an improved local outlier factor for each data point is obtained, including: The Euclidean distance between point S and point O in n-dimensional space is: ； in, For the In the historical status data of a component, the data point To data point distance; Because the health status range of each component is different; For data points Neighborhood to data point Distance is point; The improved local reachable density is calculated as: ； in, To improve the local reachability density; For data points In the neighborhood of Medium to data points No. Reachable distance; For data points The length of the neighborhood of Calculate the improved local outlier factor for each component status data: ； in, For data points Improved local outlier factor of For data points Improved local reachability density; For data points Improved local reachability density; The predicted wind turbine performance data include: ； in, The performance degradation data of wind turbines processed by the improved adaptive multi-head attention mechanism; is the connection function; Output matrix for learnable weights; The wind turbine reliability assessment model is established by using a double-layer improved nonlinear Gamma process, including: An upper-level wind turbine evaluation model is established based on the improved nonlinear Gamma process; Based on the improved raccoon optimization algorithm, the lower-level optimization model of unknown parameters of nonlinear Gamma process is established; The upper wind turbine evaluation model based on the improved nonlinear Gamma process includes: The performance degradation data of the wind turbine predicted in step 3 is Import the improved nonlinear Gamma process: ； in, , are shape parameters respectively; is a continuous nonlinear function about t; is the scale random function; is an adjustable parameter; is the scale parameter; ； in, is the Gamma function; is the probability density function of the performance degradation process of each component of the wind turbine; The distribution function is further obtained as: ； The overall failure threshold of the wind turbine is set as the time when each key component reaches the failure threshold, and the reliability evaluation model of the wind turbine is obtained. for: ； The lower-level optimization model of unknown parameters of nonlinear Gamma process based on the improved raccoon optimization algorithm includes: ； in, A random number between 0 and 1; is the location function of the unknown parameters of the nonlinear Gamma process; The value is unknown parameter set { }; is the standard normal distribution function; is the optimal position; is a random integer; M is the number of raccoons; ； Use the following method to update the location: ； in, , With the number of iterations Changing lower and upper bounds on parameters.

Images