Abstract: In order to improve the prediction efficiency of GEO electron flux greater than 2MeV one day in advance, a data fusion algorithm based on simulated annealing algorithm and least squares fitting was used to process GOES series satellite electron flux data. A genetic algorithm optimized BP neural network (GA-BP) model was established based on the fused data. The input parameters of the model include solar wind speed, geomagnetic index (including SYM/H, Ap, AU, AE, Dst), electron integral flux greater than 0.6MeV, and historical electron integral flux greater than 2MeV. The time resolution of each parameter is daily average; At the same time, using data from 1999 to 2007 as the training set, the GA-BP model after data fusion was used to predict the electron flux from 2008 to 2010, and the predicted results were compared with those of other classical models. The results showed that using simulated annealing algorithm to project satellite data located in the 75°W area to the 135°W area resulted in smaller data errors and better fusion effects; The prediction efficiency of electron flux greater than 2MeV is 0.863 one day in advance, and the highest prediction efficiency can reach 0.931, which is better than the prediction accuracy of many previous models. 

Key words: GEO satellite, GA-BP algorithm, simulated annealing algorithm, Data Fusion, high energy electron flux prediction, Deep charging