Research On Photoelectric Solar Radiation Observation Algorithm

High precision and targeted solar radiation observation data are needed in the fields of meteorology,agriculture and solar energy.However,the number of solar radiation observation stations in China is small and the distribution is uneven,which can't meet the quantitative analysis and research application based on a large number of solar radiation data.The sunshine meter based on the photoelectric principle is going to be popularized and applied throughout the country.Taking the illumination data as the main data and retrieving the solar radiation data,we can effectively compensate for the shortage of the solar radiation observation stations.However,the inversion of solar radiation is the result of multi-factor comprehensive action,which is restricted by the random conditions such as the sudden change of weather and the environmental condition.Based on the clear sky index,this paper uses PNN classification method to divide into three kinds of generalized weather types:A,B and C.The direct solar radiation is inversed by BP,GA-BP,MEA-BP and ELM algorithm under different weather types,and the results are compared and analyzed.The results show that the MEA-BP algorithm is suitable for direct solar radiation inversion of class A generalized weather type sub-models,and the ELM algorithm can be used for class B and class C generalized weather conditions.In the analysis of the influence of weather types on solar radiation,through the statistical comparison of radiation data,the classification accuracy of the PNN method is 99.4192%based on the clear sky index.For the influence of many factors of solar radiation,the PCA method is used to reduce the dimension of the multivariable,and four factors,illumination,solar height angle,temperature and humidity,are taken as the input variables of the model.In the construction of three submodels,the genetic algorithm and the thinking evolutionary algorithm are introduced to deal with the optimization of the traditional BP neural network model respectively.The results show that the determination coefficient of the MEA-BP of the three submodels reached the maximum of 0.9944.Compared with the error of the GA-BP model,the RMSE of the class A and B submodels decreased by 12.5%and 37.76%respectively,and the MABE decreased by50.47%and 55.79%respectively.Although the GA-BP error of the class C submodel was slightly smaller than that of the MEA-BP,the time consuming was it took a long time and was not universal.Therefore,applying MEA-BP method to direct solar radiation inversion can effectively improve the generalization ability.Then the ELM algorithm is introduced to inverse solar radiation and compared with MEA-BP.The results show that the RMSE is within 20W/m~2 and the MSPE is also reduced effectively when using MEA-BP method in class A and ELM method in class C.Therefore,MEA-BP algorithm is more suitable for direct solar radiation inversion under class A generalized weather type.While the ELM algorithm is much more suitable for class B and C weather conditions.