Short-term Wind Power Prediction Of Wind Farm Based On Machine Learning

The current use of wind energy is mainly through wind power integration.However,due to the volatility and randomness of wind energy,the integration of wind power will inevitably affect the stable operation of the power system.Forecasting the output power of wind farms can effectively improve the safety and stability of wind power grid connection,but its prediction accuracy is affected by the randomness of wind energy itself,while how to improve it is worth studying.Based on the historical measured data of a wind farm in the United States,based on machine learning algorithms,different predictive models and methods are used to study wind power prediction under different prediction scales.The specific research is as follows:First,in the ultra-short-term wind power prediction,since the wind power time series does not cause large mutations,many of the previous studies have used the persistence model as a benchmark model.However,the persistence model has certain limitations on the time series prediction only considering the value of the previous moment.In order to integrate more input features and improve the nonlinear learning ability of the model,the three machine learning models of support vector machine model,random forest model and Gaussian process regression model are used to study the ultra-short-term wind power prediction.The prediction effects of the three models and the persistence model and the advantages and disadvantages of the models are analyzed.The experimental results show that the machine learning model has good performance in ultra-short-term wind power prediction.Then,in the short-term wind power prediction,due to the randomness and volatility of the wind power sequence itself,the wind power prediction accuracy will inevitably decrease as the prediction time scale increases.In order to reduce the randomness of wind power sequence,combining the empirical mode decomposition algorithm with the random forest model,this paper proposes an EMD-RF combination model to improve the prediction accuracy.The original wind power sequence is decomposed by the empirical mode decomposition algorithm,and then each subsequence is predicted by the random forest model.Finally,the predicted values of each subsequence are added to obtain the final prediction result.The experimental results show that the prediction accuracy of the proposed combined prediction model is significantly improved compared to the three separately used machine learning models under the one-hour prediction scale.Finally,in the short-term multi-step wind power forecasting,in order to better learn the trend of wind power sequence,a method based on the combination of wind speed similar day clustering and machine learning model is proposed in this paper.Only the wind speed value with the highest correlation with wind power is selected as the feature.The sample days with similar wind speed changes are clustered by the improved K-means algorithm,and then the model is trained.Then,the wind speed day category to which the forecast date belongs is judged and passed.The corresponding model was subjected to four consecutive six-hour multi-step predictions.The experimental results show that the screening of training data by wind speed similar day clustering can effectively improve the accuracy of multi-step wind power prediction.