Study On Wind Turbine Generator Power Performance Degradation Based On SCADA Data

In recent years,wind power generation has developed rapidly in China.Realtime monitoring of the power generation performance of wind turbines is of great significance for improving the safety and economic efficiency of the units.The SCADA data recorded during the normal and stable operation of wind turbines,namely the power main band data,can accurately reflect the power generation performance and data characteristics of wind turbines.However,in actual operation,due to factors such as artificial load limitation and sensor failure,there are often a large number of abnormal values in the measured operating data,which cannot be directly used for the study of power generation performance degradation.Therefore,this paper conducts research from two aspects:wind turbine abnormal data identification and power generation performance degradation warning.(1)Analyze the control strategy and data characteristics of wind turbines under different operating conditions.Starting from the wind speed-power scatter plot,the distribution of power main band data and abnormal data of wind turbines is demonstrated,and the main causes of abnormal data generation are studied to provide a basis for subsequent modeling.(2)A wind turbine abnormal data identification and power generation performance degradation monitoring method based on the DB-MKIF model is proposed.The identification of abnormal data in wind turbines is the basis for power generation performance degradation warning.The DB-MKIF model uses an adaptive DBSCAN clustering algorithm to preliminarily remove abnormal data between partitions.Then,the MKIF algorithm based on isolation forest is proposed to define the MKIF anomaly score and identify and remove global abnormal points in the running data in three-dimensional space to extract the power main band data composed of normal data and calculate the rejection threshold.The missing values in the monitoring data are filled using the KNN algorithm.In the monitoring phase,the MKIF anomaly score of the monitoring data is calculated,and the data with anomaly score greater than the rejection threshold is marked as degraded data.The degradation rate of time series data within the monitoring window is monitored using the sliding window method,thereby achieving power generation performance degradation warning of wind turbines.Using actual operating data as an example,the DB-MKIF model can accurately identify abnormal data,determine the degradation of wind turbine power generation performance,and identify abnormal gearbox oil temperature.(3)A regression prediction-based wind turbine degradation monitoring method using SHAP feature selection and XGBoost modeling is proposed.In order to monitor the working state of wind turbines in more dimensions,a regression prediction method is used to establish a data model and monitor the degree of deviation of relevant variables from the model,thereby achieving power generation performance degradation warning of wind turbines.By analyzing the control strategy and data characteristics of wind turbines,monitoring variables under different operating conditions are determined.The SHAP algorithm is used for feature selection,and the XGBoost algorithm is used to establish data models of various monitoring variables of wind turbines under different operating conditions,with the power main band data as input.Then,the sequential probability ratio test method is used to analyze the prediction residual.When the prediction residual of the model undergoes abnormal changes,a power generation performance degradation warning is issued.This paper verifies the effectiveness of this power generation performance degradation warning method using a set of blade icing data.