Solar Radiation Synergy Prediction Based On Machine Learning

With the world energy crisis becoming increasingly apparent,the development and application of clean energy has attracted more attention.Photovoltaic power generation is one of the main ways to use solar energy.Since solar radiation is the main factor that determines the power generation of photovoltaic systems,solar radiation prediction has become the basis of the output power prediction of photovoltaic power generation systems.Accurate hourly radiation data can quantify the impact of solar radiation on buildings,lay the foundation for building dynamic energy consumption simulation and building energy conservation design,and also improve the utilization rate of solar power generation.At present,most research on the solar radiation prediction only considers the data of the target station to be predicted.In order to improve the prediction accuracy of solar radiation at the target station,this thesis proposes a synergy prediction model,which takes the data of neighbor stations into account,and uses the data of seven stations in northern Brazil for experiments.The main work is as follows.This thesis introduces the data set used in the experiment and the basic knowledge of data preprocessing,completes the missing data and codes the month and hour with sine function,which provides the basis for the following prediction models.Three evaluation indicators are presented to compare the performance of the proposed models.When considering only the historical solar radiation values of the target station and neighboring stations,the EEMD-GRU-SVR synergy prediction model is proposed to predict with one step in advance the target station.The ensemble empirical mode decomposition(EEMD)is firstly used to decompose the solar radiation data of the target station into multiple components,then the gated recurrent unit(GRU)is adopted to predict each component,and the prediction results are finally summed to obtain the preliminary prediction values.This method takes the preliminary predicted value and the historical solar radiation data of neighboring stations as the input of the support vector regression(SVR),and the corresponding output is the final predicted result.When predicting the solar radiation of the target station,the meteorological and time characteristics of each station are also considered besides the solar radiation values of the target station and the neighboring stations.A short-term solar radiation synergy prediction model is proposed based on the maximum information coefficient(MIC).First,the maximum information coefficient is used to perform feature selection on relevant data from all stations.Then the data after feature selection are utilized as input for prediction using different machine learning models.The errors of the synergistic prediction is finally compared to that taking only the target station data on real data.Experimental results indicate that the synergy prediction has a higher precision and low error for all target stations,compared the methods without synergy prediction.In order to further improve the prediction performance,a synergy prediction method of solar radiation is established based on MIC-RF-LSTM.Firstly,features are selected by MIC according to the data from all stations.Then the random forest(RF)model is utilized to forecast the solar radiation of the target stations based on all selected features.Finally,the prediction results are further corrected by the long short-term memory(LSTM)model.The experimental results validate that the proposed hybrid prediction method can significantly enhance the solar radiation accuracy.