Wind Farm Power Prediction Based On Machine Learning

With the depletion of fossil fuel and the deterioration of human living environment,the development of renewable energy has become the focus of attention around the world. Wind energy which is clean and widely distributed is thus paid more and more attention. Our country is rich in wind energy resources and many large-scale wind power farms are built in recent years. However, due to randomness and fluctuation,integration of wind power to the grid brings with it serious issues which greatly restrains its large scale development. To mitigate the ha rm to the grid caused by the fluctuation of wind power and ensure stable operation and appropriate distribution of power grid,the research on the wind power prediction becomes particularly important.This dissertation focuses its research on the establishment of wind speed forecasting model and wind power forecasting model through the use of Machine learning, statistical analysis and other technologies, based on which the wind speed and wind power of wind fram can thereupon be predicted. The following aspects are mainly studied:Firstly, random forest algorithm is adopted for Prediction of wind speed. With the historical wind speed data as the sole training sample, the random forest prediction model is established, which leads to the possibility of single step prediction of wind speed in the next one hour. The results prove the feasibility of wind speed prediction through use of random forest.Secondly, two methods of single step prediction of wind power in the next one hour are studied. With the historical wind power data used as the training sample,prediction model can be established by using the TreeBagger function in MATLAB,and meanwhile "predict" function can be used to forecast future wind power. Then a method of predicting wind power by using of random forest algorithm is proposed and the random forest prediction model is established. After analyzing and comparing the effects of different algorithm parameters on the proposed model, satisfying predictionresults are obtained.Lastly, 3 methods are used to predict the wind power in the next four hours and prediction results are compared. First ,the time series forecasting algorithm ARIMA is used to predict the wind power in the next 4 hours, and the optimal parameters of the ARIMA algorithm are selected. Then empirical mode decomposition (EEMD) and the TreeBagger function in MATLAB are used to carry out the short-term multiple-step rolling forecast of wind power, which shows the forecast curve of the wind power in the next four hours. In the end, the method of using EEMD and random forest algorithm to forecast the wind power is described. Comparing the prediction results of these 3 methods, it is shown that the random forest algorithm can more accurately predict the short-term wind power.Wind power forecasting is an important research topic. Due to the randomness and intermittent features of wind power,wind power forecasting is also a tough issue to handle. The random forest algorithm which is a new machine learning algorithm,has some unique advantages and research on using it to forecast wind power has just started.This dissertation conducts some exploration on this topic and is expected to provide some help for further wind power forecasting work.