Research On Short-term Wind Power Prediction Based On Ensemble Learning

With the rapid development of the world economy,the consumption of traditional fossil energy not only cause a global energy crisis,but also leads to serious deterioration on natural environment.Wind energy,as a clean and renewable energy,has received wide attention from both domestic and overseas.However,the strong volatility of wind energy and the grid connection process of wind power will seriously threaten the security and stability of the overall power grid.Based on accurate prediction of generated power of wind energy,the generated power efficiency can be effectively improved and can largely reduce the threat of grid connection process to power grid.This paper takes short-term wind power prediction as the research object.In order to make better use of the historical data of wind,based on the detailed analysis of the historical data of wind power in a wind farm,this paper mainly completed the following research work:(1)Firstly,this paper introduces the research progress of wind power prediction at home and abroad,and analyzes the importance of wind power prediction in detail,which leads to the significance of wind power prediction research.(2)The wind power under different meteorological conditions usually has a large difference,so the model should be trained similar data.This paper proposes a similar day selection strategy based on historical data clustering analysis and K-nearest neighbor algorithm evaluation,which effectively improves the prediction accuracy of the model.(3)The fluctuation of wind energy increases the difficulty of model prediction.Single model usually has some shortcomings such as insufficient generalization ability and poor robustness.Ensemble learning model effectively compensates for these shortcomings,so it's an appropriate model for wind power prediction.Based on the introduction of ensemble learning concepts,this paper discusses how to improve the effect of ensemble learning from three aspects: individual learner generation,ensemble pruning and ensemble strategy.(4)In order to improve the accuracy,generalization and robustness of the prediction model,an ensemble learning algorithm based on dynamic weighted fusion is proposed.In this paper,a strategy for generating individual model training set is proposed,which combines Bagging algorithm with random feature subset selection algorithm based on roulette.The integrated learning model uses the improved BP algorithm as the individual model training algorithm.This individual model training strategy not only ensures the accuracy of individual learners,but also effectively increases the difference between individual learners.In the integrated pruning process,this paper uses the clustering-based integrated model pruning strategy to select the optimal and the most different individual models as the base learner of the ensemble model.Because PSO algorithm has good optimization performance,this paper uses PSO algorithm to find the optimal combination coefficient based on similar samples,and uses it as the dynamic weighting coefficient of integrated learning.(5)In order to verify the effectiveness of the algorithm,this paper compares the prediction results of the model with some common wind power algorithms.It is shown in experiments that the ensemble model effectively improves the prediction accuracy of wind power and has certain practical application prospects.