Research Of Short-Term Wind Power Forcasting Based On Cloud Computing And Machine Learning

With the adjustment of China's energy structure and the rapid growth of installed capacity of Wind Power(WP),timely and accurate prediction of wind power can provide an important basis for rational management of th e power grid,and improve the utilization of wind power effectively.At the same time,with the improvement of the wind farm intelligence level,the wind power monitoring data scale is increasing,which poses new challenges to the calculation performance o f traditional wind power forecasting model.In recent years,artificial neural network method and support vector machine method based on machine learning theory and its improved algorithm have been widely used in short-term wind power forecasting.There are many iterations in the machine learning algorithm.Spark,a distributed memory computing framework in cloud computing,can efficiently process iterated data and improve the execution performance of the algorithm.For the existing short-term wind power forecasting model,there are some problems such as generalization is weak,the model structure and parameters are difficult to determine,and the interpretability is poor.In this thesis,a short-term wind power forecasting method based on improved random forest regression algorithm is proposed by combining random forest regression algorithm,M5 P model tree,differential evolution algorithm and selective integration method,and the algorithm is parallelized by using the Spark cloud computing platform.The main work of this thesis is as follows:(1)The traditional random forest regression algorithm takes the classification regression tree as the meta decision tree.The prediction accuracy of classification regression tree is low,and it can not give a continuous output and the prediction value can not exceed the range of training set data.In this thesis,the M5 P model tree is used as the meta-decision tree of the random forest regression model.By constructing the multiple linear regression model on the leaf nodes,the prediction accuracy of the meta-decision tree is effectively improved.(2)There are some meta-decision trees with poor predictive performance and low diversity in random forest.In this thesis,an improved differential evolution algorithm is proposed and applied to the selective integration of random forest decision trees.A new random forest regression model is constructed by selecting a subset of the optimal meta-decision trees in all the meta-decision trees,and the final results are calculated by weighted calculation.(3)Aiming at the problem of high computational complexity of random forest algorithm,the parallelization method of random forest algorithm and differential evolution algorithm is analyzed.By studying the architecture of cloud computing system and adopting the Spark distributed memory computing framework in cloud computing technology,the above-mentioned prediction algorithm is improved in parallel to improve the execution performance of the algorithm.(4)In this thesis,wind power monitoring data of a certain area in Inner Mongolia is taken as a practical example.The method proposed in this thesis is compared with the existing short-term wind power forecasting algorithm and the traditional random forest regression algorithm.In addition,Cloudera's CDH5 version is used to build the cloud computing platform on the laboratory server,and the performance of the proposed algorithm is tested.The experimental results show that the method proposed in this thesis has high prediction accuracy,generalization performance and interpretability,and has good parallel performance.