Study of improved genetic algorithm based on neural network

Abstract

In order to overcome these disadvantages such as low rate of convergence in neural network back propagation (BP) algorithm, the likeliness to fall into local minima, the absent foundations for selecting initial weight values and threshold values as well as great randomness, the neural network optimization method is developed based on adaptive genetic algorithm. This technique combines advantages on both neural network BP algorithm and neural network optimization method. However, it proves to be unsatisfactory due to the slow evolution of population in early stage. The proposed algorithm is applied in the optimized design of weight values and threshold in neural network. The algorithm model is employed to make forecasts about unfathomable SVI parameters in a sewage plant in one city. The simulation experiment suggests that the aforesaid algorithm can not only eliminate shortcomings in BP algorithm, but also can improve remarkably convergence speed and precision through the comparison of its model with BP and GA-BP network models, obtaining good results. © 2012 Springer-Verlag Berlin Heidelberg.