Health Status Evaluation Of Wind Turbines Based On SCADA Operation Data

As a clean renewable energy source,wind energy is increasingly valued for its technical level,infrastructure construction and cost advantages.However,the volatility,intermittentness and low energy density of wind power also cause the volatility of wind power,which is easy to have a huge impact on the stable operation of wind turbines and grid interconnection.Therefore,the evaluation of the health status of wind turbines is very Significance.In this paper,the wind turbine health assessment based on SCADA operation data is studied.According to the two different research directions of nuclear model and neural network model,Select SVR model based on interval clustering and CNN model based on LSTM.SVR is a kernel learning algorithm that uses support vector machine to fit curves and perform regression analysis.This paper proposes the idea of interval clustering for wind turbine data and SVR characteristics: The unit operation data is divided into sections according to the wind speed,and the center point of each section is found by clustering method as the training data of the SVR model.Among them,the interval length and the number of center points use the particle swarm optimization algorithm to find the optimal solution,which avoids data loss and precision degradation caused by interval clustering.Cyclic neural network can process arbitrary length information according to time series.LSTM can selectively forget and memorize historical data.This paper combines the characteristics of both to establish a LSTM-based cyclic neural network model.The accuracy of the wind turbine model using this method is verified by the example to be higher than that of the traditional neural network.For the nuclear model and neural network model,this paper proposes a predictive degradation degree index,and combines the SCADA data of the same wind farm to evaluate the health status of the unit.It is verified that the SVR model based on interval clustering is more accurate,the LSTM model is more adaptable.