| With the economy developing of our country,pollution crisis caused by traditional energy consumption is becoming more and more serious.In order to alleviate the trend,the new energy development and application have been attached exactly attention.Because of advantages of clean,safe,renewable and inexhaustible,solar energy has received great attention.In order to improve the efficiency of solar energy,it is extremely important to predict solar radiation situation.This thesis,aimed at studying the solar radiation prediction problem further,combined with advanced machine,effectively predict solar radiation through the random forest regression prediction function in theory,provides a novel method for the solar radiation application in the photovoltaic power generation,which has important research value and broad application prospects.In this thesis,the main work is arranged as following:(1)After consulting relevant technical literature at home and abroad,in the thesis,the present solar energy utilization situation and the development trend of solar radiation prediction method has been summarized,the relative factors of solar radiation have been analyzed,the main influencing variables of the solar radiation has been determined,the basic solar radiation mathematical model has been built.(2)After analyzing the basic principle of solar radiation and the related factors,the main influence factors have been determined,the factors including the angle of the sun,sunshine time,clear sky index and so on.The characteristic curve of solar radiation and influence factor is obtained by the simulation,the analysis results show that the longer the sunshine time,the greater the solar radiation.With the solar altitude angle increasing,the distribution area of the same amount of solar radiation will be increasing,so does the solar radiation intensity per unit area.(3)The random forest method of solar radiation prediction has been research.A random forest tree regression classifier has been established with the nine main variables.The same region but different periods solar radiation data are been used to get the correlation radiation factor through simulation experiment and build the solar radiation prediction model based on random forest.(4)The studying compare with different prediction methods,the solar radiation prediction model based on the BP neural network and the Support Vector Machine(SVM)have been respectively established.Then the predictive result of the above two models has been compared with the result of random forest prediction method under the condition of the same training samples.The experiment result show that the predictive effect of random forest model is better,and the root mean square error reduced and the forecast precision of model improved,when estimate the solar radiation amount of no man’s land.(5)The improved solar radiation method has been studied.For the unbalanced data caused by the more random disturbances of solar radiation in urban areas,a random forest algorithm based on C_SMOTE algorithm has been put forward.The simulation researching of random forest algorithm before and after the improvement is carried out.It can be seen that the OOB error is reduced and the validity of the improved random forest method is verified by comparing the predicted results with other methods.Simultaneously,it is of great significance and application prospects to forecast the solar radiation amount in complex environment. |
