Application Of Solar Radiation Estimation Model Based On Machine Learning And Data Fusion

Solar energy resource is a kind of high quality renewable energy,is an important part of clean energy.China has proposed a two-carbon goal of peaking carbon emissions by 2030 and achieving carbon neutrality by 2060.Solar radiation is closely related to solar energy resources,so studying the change and distribution of solar radiation can help us effectively utilize solar energy resources.At the same time,solar radiation can support agriculture,hydrology,meteorology and other fields related research.However,in China and many other countries around the world,there are few and sparse radiation observation stations,which can be effectively solved by satellite remote sensing data.In addition,the use of FY-4A satellite data products in the study of solar radiation is relatively rare.In view of the above situation,the specific research work of this paper is as follows:(1)FY-4A data was fused with ground station data to construct experimental data set.First,7 kinds of data products of FY-4A were selected for reading,conversion,tailoring,cleaning and other operations,and then matched with the data of five meteorological stations in Guangxi by means of latitude and longitude distance,and 39,766 pieces of data used in the experiment were obtained.Secondly,the mutual information method is used to analyze the correlation between FY-4A stationary satellite data and the measured data of five radiation observation stations in Guangxi.According to the correlation,cloud top height(CTH),cloud top temperature(CTT),total atmospheric water vapor(LPW),surface temperature(LST),cloud coverage(CFR),observation DAY(day)and observation HOUR(hour)were selected as the input of the estimation model.(2)A solar radiation estimation model based on the integrated learning strategy of Stacking is proposed.Firstly,a comparative analysis of the effects of five machine learning algorithms in solar radiation estimation is carried out to verify the effectiveness of the integrated learning method in solar radiation estimation.Moreover,the error of solar radiation estimation is low in spring and winter,and high in summer and autumn,especially the error in May-September is much higher than that in other months.Secondly,considering the historical information of time series,the Recurrent neural network(RNN)and long and short term memory network(LSTM)were used to construct the solar radiation estimation model,which obtained better estimation results than the five machine learning methods.Finally,based on the Stacking integrated learning strategy,the optimal RF algorithm of five machine learning methods is combined with RNN and LSTM to construct the integration model.The results show that the integrated model has the best performance.(3)Research and development of solar radiation data estimation and monitoring system based on model research results.The system includes data management,data processing,data storage,estimation model,data statistics and data display modules,which can show the solar radiation distribution in Guangxi and facilitate the application of model estimation results to related businesses.