Research On Wind Power Forecasting Method Based On Hybrid Data-driven Model

As a renewable energy with great development potential,wind energy has the advantages of wide resource distribution,high conversion efficiency and low environmental impact.The development and utilization of wind energy have received increasing attention.However,when large-scale wind power is connected to the grid,its intermittency and randomness will bring great challenges to the stable operation of the grid and the quality of power supply.Accurate prediction of wind power can significantly improve the safety,stability and economy of the power system.Therefore,research on wind power forecasting methods have important practical significance.Based on actual wind farm data,wind power forecasting methods are studied by using data-driven technology and swarm intelligence optimization algorithms.The main research work includes:(1)The original wind power data are pre-processed accordingly.First,the wind power data collected by a wind farm in Yunnan and a wind farm in Turkey are taken as the original data set A and data set B,respectively.For the original data set A,the recursive feature elimination(RFE)method is used for feature selection to find the optimal feature combination.For the original data set B,the quartile cleaning method and the model filling method are used for preprocessing.(2)A hybrid wind power forecasting model is proposed by combining extreme learning machine(ELM)and improved teaching learning based optimization(i TLBO).In order to improve the convergence speed and learning ability of the basic TLBO algorithm,four improvements are made.The obtained i TLBO algorithm is applied to the parameter optimization of the ELM model.Wind power forecasting results show that the hybrid forecasting method has good accuracy.The relative root mean square error,average absolute error and average absolute percentage error are all lower than the given comparison methods.(3)An ensemble wind power forecasting method is proposed by partial least squares method.Three base models with different working principles are selected in this ensemble forecasting method.In order to improve the prediction accuracy and reduce the error of a single model,a PLS method combination mechanism is used to set the weights of these base models.Wind power forecasting results show that the proposed ensemble method can combine the advantages of each base model and effectively improve the prediction accuracy.