Construction Of Short-term Solar Radiation Prediction Model Based On Firefly Optimization Algorithm

Faced with the deterioration of ecological environment and the continuous depletion of traditional energy,the development of clean energy has become an urgent demand.Among the numerous clean energy,solar energy has become the mainstream of renewable energy due to its extensive distribution,green environmental protection,abundant resources and other advantages.In recent years,with the development of solar power generation industry,more and more researchers pay attention to the application of various prediction models in solar power system.In this paper,the Lasso-FA-SVR model and the Lasso-FA-BP neural network model are proposed.Since there are too many factors affecting solar energy prediction,the feature selection method Lasso is integrated with solar energy prediction model in this paper,and the support vector regression model and BP neural network model are improved by combining firefly optimization algorithm.This model can not only reduce the dimension of the feature vector of the raw data,but also optimize the related parameters of the model so as to improve the learning performance.In this paper,the model is successfully applied to short-term solar radiation prediction,and the prediction results of the proposed model is compared with that of FA-SVR and FA-BP neural network model.In this paper,the proposed model was experimented on short-term solar radiation prediction for many times,and the research results showed that the prediction accuracy of Lasso-FA-BP neural network model is the highest among all models.And compared with FA-BP neural network model,MAE average decreases by 22.61%,RMSE average decreases by 16.52%,MAPE average decreases by 22.03% and the training time of the model is average shortened by 38.13%.However,the prediction accuracy of Lasso-FA-SVR model is not improved compared with FA-SVR model,but the training time of the model is average shortened by 86.13%.Therefore,Lasso-FA-SVR model can be selected for prediction when the accuracy of the model is not very high.