Simulation Of Solar Radiation And Study On Temporal-spatial Changes In Northwest China Based On Neural Network

As the basic power of the climate formation and evolution, solar radiationfundamentally determines the energy balance of the earth’s atmosphere. Solarradiation data can be used to simulate the surface dynamic and thermodynamicprocesses. It is important input parameters of many models, such as ecology model,hydrology model, crop model, radiative transfer model and global circulation model.The quantitative estimation of solar radiation is very important to learn andunderstand arid and semi-arid region climate change. The distribution of radiationstations in Northwest Region is rare and uneven. Therefore, the radiation data thatobtained from interpolation or extrapolation through scarce radiation sites areuncertain. BP (Back propagation) neural network model has good performance in theprediction of solar radiation. However, previous studies of estimating solar radiationare based on some single sites. We can not obtain spatial continuous solar radiationdata through rare sites. But BP neural network model also has many inherentshortcomings in practical applications, such as long learning time and slowconvergence speed. In order to overcome these shortcomings, an attempt has beenmade to investigate its application possibility with LM (Levenberg-Marquardt)arithmetic combining neural network. The paper adopted the improved LM-BP neuralnetwork model to estimate the monthly global solar radiation with meteorological datain Northwest Region during1990–2012. Monthly global solar radiation values areestimated for159out of184stations using LM-BP neural network model, measuredvalues of monthly global solar radiation are used for the remaining25stations. Thevalues for184stations are then utilized for the interpolation in Northwest Region,then, analyzes the spatial-temporal variation characteristics of global solar radiation.The research results can provide important theoretical references and data support forknowing the variation tendency of global solar radiation and developing various kindsof models. In addition, it also provides a good method of calculating the solarradiation for some areas which has no radiation observation. The main researchcontents are as follows:(1) BP neural network model can well deal with nonlinear problems that the general regression models can’t solve, but serious redundancy of input data makes thenetwork complex and unsteady. So we use the connection weights method to simplifythe network structure. Through computing the relative contribution of each inputvariable, we can estimate the importance of each input variable to the output variable.The redundant input variables can be eliminated and the network structure can besimplified. Then, the convergence speed and learning efficiency of the network is alsoimproved.(2) Using MATLAB platform, we adopt the improved LM-BP neural networkmodel to estimate the monthly global solar radiation of Urumqi, Kashi, Hami, Xining,Guyuan stations in Northwest Region, combining the measured data, compared withother two empirical models using the precision test and error analysis.(3) Using the improved LM-BP neural network model to estimate the monthlyglobal solar radiation in Northwest Region during1990–2012, the simulative valuesand the measured values are utilized for the interpolation in Northwest Region. Thenwe analyzes the spatial-temporal variation characteristics of solar radiation. Finally,we analyze the reasons that result in the change of global solar radiation.Through the study of this article, the main conclusions are as follows:(1) The improved BP neural network with LM algorithm has better ability ofnonlinear mapping than the traditional BP algorithm. It obviously shortens the trainingtime, reduces the training error and shows a good fitting performance. LM-BP neuralnetwork model has more rapid convergence speed and higher precision than other twoempirical models through the verification of five stations. The results indicate that theimproved BP neural network is the most effective on estimating solar radiation inNorthwest Region, the error indicators is the smallest among the three models. Thecorrelation coefficients for simulative values and measured values of five stations areover0.94.(2) Through the spatial interpolation analysis, it is concluded that the averagemonthly global radiation in Northwest Region changes between262MJ/m2and643MJ/m2, solar radiation is rich in central region and less on both sides. Monthly globalsolar radiation showed significant decreasing trend from1990to2012in NorthwestRegion, the tendency rate changes between–88.9MJ/(m2·10a) and30.1MJ/(m2·10a). During1961–2012, annual global solar radiation presented obvious decreasing trend,averaged tendency rate is–74.8MJ/(m2·10a). The change of annual global solarradiation has experienced two stages: a significant decrease, the tendency rate is–237MJ/(m2·10a) before1990, however, global solar radiation has been increasing slowlysince1990, and the tendency rate is14MJ/(m2·10a).(3) The average monthly global radiation range of the four seasons is between170MJ/m2and830MJ/m2during1990–2012, the annual spatial distribution ofmonthly global solar radiation shows that the locations and range of the central highvalue area and the western low value area have a great changes, however, the easternlow value area has few changes.(4) According to analyzing of the causes of global solar radiation changes, wefound that it includes sunshine duration, cloud cover, aerosol, water vapor, temperature,etc., the sunshine duration has significant positive correlation with the global solarradiation, the effect of cloud cover and water vapor on the global solar radiation is notsignificant, the aerosol is the main factors influencing its change.