A Comprehensive Application Of Machine Learning Techniques For Solar Radiation Prediction

With further improvement and rapid development of photovoltaic(PV)power generation technology system in China,PV generation technology has gradually become the dominant trend.Large capacity and centralized grid-connection make PV power generation system capacity get rapid growth,but the maximum output power generated by centralized grid-connected PV system is still an intermittent growth and uncontrollable in essence.These inherent characteristics have an adverse impact on the power grid and seriously restrict grid-connected PV power generation.Forecasting the output power of solar PV systems is required for the good operation of the power grid and the optimal management of the energy fluxes occurring into the solar system Before forecasting the solar systems output,it is essential to focus the prediction on the solar irradiance.In this thesis,the solar radiation data collected at a certain place in Jiangsu of China for two years are investigated.The main results are as followsCombining a novel clustering technique with unsupervised machine learning,a solar radiation data preprocessing method is proposed.Firstly,the missing data is recovered by matrix completion method.Then robust principal component analysis is used to de-noise the data after completion.In order to reduce the influence of weathe types on solar radiation,a spectral clustering method combining k-nearest-neighbo representation and sparse subspace representation is used to cluster the data s etFor the clustered data sets,different neural network models are employed to predict the solar radiation,and their performances are compared.The experimental results show that the proposed solar radiation prediction method does improve the prediction accuracy as a whole,and reasonable data preprocessing and sample division have a beneficial impact on the short-term solar radiation predictionAn integrated learning model is designed to predict PV power generation by combining various factors.The model takes solar radiation,sunshine temperature and power consumption as input variables,which can improve the prediction accuracy of power generation to a certain extent.