A fast BP algorithm of combined global optimization based on optimizing initial weights
DONG Guang ping, FANG Min, WANG Jing wei (School of Electrical Engineering, Hefei University of Technology, Hefei 230009, China)
Based on a lot of experiments, the slow convergence of BP algorithm is deeply researched. For realizing a fast global optimization of forward network training, the variable step method is combined with simulated annealing (SA) method based on optimizing initial weights. The new algorithm is applied in system modeling,which shows that the suggested algorithm has fast convergence and high precision.
【CateGory Index】： TP183