Simulation Of Solar Radiation Based On Neural Network In Eastern China

Climate change is a major global issues of common concern of the international community,over the past century,the earth is experiencing a temperature rise,while solar radiation is an indicator of climate change.At the same time,Solar radiation data is important parameters about crop models,hydrological models and climate change models.The quantitative simulation of solar radiation is essential for us to learn and understand global and regional climate change.The distribution of radiation stations in global is rare and uneven.Therefore,the radiation data that obtained from interpolation or extrapolation through scarce radiation sites are uncertain.Thus,the lack of solar radiation data limits the research in related fields.As we all known that there are many uncertain factors that affect the change of solar radiation,and the influence factors will affect each other.BP?Back Propagation?neural network model has strong nonlinear processing ability,However,previous studies of estimating solar radiation are based on some single sites.We can not obtain spatial continuous solar radiation data through rare sites.But BP neural network model also has many inherent shortcomings in practical applications,such as long learning time and slow convergence speed.In order to overcome these shortcomings,an attempt has been made to investigate its application possibility with LM?Levenberg-Marquardt?arithmetic combining neural network,this algorithm has fast local convergence feature about Gauss-Newton method,but also has global search feature about gradient descent method,which allows error along the direction of deterioration to search,and greatly improving the convergence rate and generalization ability of the network.But existing neural networks rarely consider cloud,aerosol,and perciptible water vapor influence on solar radiation.In this paper,Firstly,We comprehensively analyzed the effects of various factors on solar radiation.Secondly,Based on LM-BP neural network,we used cloud,aerosols,atmospheric precipitable water vapor from MODIS atmosphere remote sensing products and conventional meteorological data to simulate solar radiation.Because of the good quality of data in Eastern China,in this paper we used Eastern China as the experimental area to verify the solar radiation simulation method.In this paper,the monthly mean of solar radiation of 90 conventional weather stations was simulated by using the LM-BP neural network model.The input parameters of the model were latitude,altitude,aerosol optical thickness,cloud fraction,cloud optical thickness,atmospheric precipitable water vapor,sunshine hours,air pressure and air temperature about Eastern China from 2001 to 2014.We validated the model with the measured values.The model had the best fit of 0.95,the root mean square error was only 0.57 MJ·m^-2,and the average deviation error was basically-1 MJ·m^-2 to 1 MJ·m^-2.This proved that the simulation accuracy of the model was high.Finally,combining the measured values of 13 radiation stations and by spatial interpolation,we got the spatial distribution of monthly mean solar radiation,and we also analyzed the temporal and spatial distribution characteristics of solar radiation.Through the study of this paper,some general conclusions could be drawn as follows:?1?There was a positive correlation between sunshine hours,wind speed and solar radiation,and there was a negative correlation between air pressure,aerosol,precipitation water vapor,cloud optical thickness,cloud fraction and solar radiation.On the whole,the influence of various factors on solar radiation in different areas was different.?2?When solar radiation reaches the surface through the atmosphere,clouds,aerosols,precipitation water vapor and other atmospheric factors on the impact of solar radiation is great,in the simulation of solar radiation should consider these factors.?3?In this paper,we established a neural network simulation model with different structures and compared the simulation values and the measured values.It was found that on the basis of latitude,altitude,sunshine hours,air pressure and air temperature added aerosol optical thickness,cloud fraction,cloud optical thickness,precipitation water vapor,the simulation accuracy was the highest,and the goodness of fit was up to 0.95,the root mean square error was controlled within 2 MJ·m^-2,the average deviation error was between-1 MJ·m^-2 and 1 MJ·m^-2,the average percentage error was basically controlled within 10%.?4?Combining MODIS with conventional meteorological data to simulate the solar radiation based on LM-BP neural network is a good solar radiation simulation method,which is suitable for simulating solar radiation in sparse radiation stations areas.?5?The change rate of surface solar radiation in eastern China was between-0.06 and 0.18 in during 2001²014,and the whole trend was increasing;the monthly average of solar radiation in Eastern China was between 7.02 MJ·m^-2 and 23.89 MJ·m^-2 in the period from January to December of 2001 to 2014,being rich in both sides and less on central region.The range of the monthly average solar radiation in the four seasons was between 7 MJ·m^-2 and 24 MJ·m^-2,From the whole distribution,the solar radiation was the strongest in summer,followed by spring,Autumn,winter with the weakest solar radiation.