Ultra Short Term Prediction Of Wind Power Based On Machine Learning

Since the industrial revolution,the world has made great efforts to develop industry.In recent years,China has focused on the transformation from a big manufacturing country to a powerful manufacturing country.With the development of economy,the demand for all kinds of energy is also growing,and the environmental pollution caused by energy consumption is becoming increasingly serious.Therefore,it is urgent to develop clean energy to replace non clean energy.Wind energy,as an important clean energy,has been favored by all countries.While putting forward the goal of building smart grid,China emphasizes the important role of wind power generation,and the accurate prediction of wind power generation affects the construction and upgrading of smart grid to a certain extent.Therefore,in order to better help wind power enterprises accurately predict wind power generation and provide reference for the development of smart grid In this thesis,the ultra short term prediction of wind power generation is studied.This thesis mainly uses two methods to reduce the data dimension,which are principal component analysis and maximum information feature selection;and four machine learning regression methods as prediction models,which are decision tree regression,random forest,support vector machine regression and neural network Then,two dimensionality reduction methods and four machine learning methods are compared.It is concluded that feature selection method has higher dimensionality reduction efficiency and better effect than principal component analysis method in ultra short term prediction of wind power,which is more suitable for the research content of this thesis.It is also concluded that the accuracy R2 of random forest method can reach 0.82 under the premise of data in this thesis,which is correct Among the four methods,the most suitable one is the conclusion of the model for wind power prediction,and the combined method of feature selection and random forest is fast and good.After that,Tpot method is selected to optimize the model from another angle,and another set of data is used to fit the feature selection random forest method to test its applicability.It is concluded that this method has certain practicability in other data,which can provide reference for the output power prediction value after 15 minutes.At the end of the thesis,according to the research conclusions and development status,some suggestions for better construction of smart grid are put forward.