Research On Ensemble Model For Short-Term Wind Power Forecasting

The power system dominated by renewable energy is an inevitable trend towards carbon neutrality and peak carbon emissions.Wind power,as a clean new energy source,has developed rapidly in recent years.However,its dependence on uncertain meteorological conditions leads to randomness and fluctuation of wind power,which brings great challenges to the safety,economy,and stable operation of the power system.Therefore,accurate wind power forecasting has important practical significance and application value.Using actual wind farm data,this thesis focuses on hybrid wind power forecasting methods based on ensemble learning strategy.The main work includes:(1)Wind power data preprocessing methods.The sources and characteristics of wind power data were analyzed.The local outlier factor(LOF)algorithm and quadrature method were used to detect and remove abnormal data.The back propagation neural network was used to interpolate the removed data.Based on the daily similarity of wind power data,K-means clustering was performed on the training set and then K-nearest neighbor(KNN)algorithm was utilized to categorize the testing set.These preprocessing methods lay the foundation for the subsequent forecasting model research.(2)Wind power forecasting based on selective ensemble strategy.Following the principle of low correlation and high forecasting accuracy,four base models with different internal mechanisms were selected from the model library,namely,the improved teaching-learning-based optimization extreme learning machine(i TLBOELM),adaptive neuro-fuzzy inference system(ANFIS),relevance vector machine(RVM),and Gaussian process regression(GPR).Results of each base model were combined using partial least squares regression(PLS)to obtain the final forecasting results.Comprehensive comparison and analysis demonstrate the effectiveness and adaptability of the proposed forecasting method.(3)Wind power ensemble forecasting based on data clustering.Based on the research work(1),the training set and testing set were subjected to K-means clustering and KNN classification.Based on the research work(2),some base models were added and optimized.Finally,the ensemble learning strategy was applied to the model training for each cluster.Results show that this forecasting method can effectively improve the wind power forecasting accuracy.In addition,with the increase of the forecasting horizon,the accuracy improvement brought by clustering becomes more evident.