Neural Network Trained By Bat Algorithm And Its Application

Because of its strong nonlinear mapping ability, the Artificial Neural Networks are widely used in the field of numerical prediction and engineering control. But its global search optimization ability is poor and traps into local optima easily. Therefore,the data forecasted by the technology of neural network is not accurate. Bat algorithm is a kind of intelligent optimization algorithms which is widely concerned at present. Bat algorithm has been applied widely due to its simple arithmetic, fast convergent rate and good global search ability. By analyzing and summarizing the mechanism of the BP network and the bat algorithm, a cooperative neural network learning method based on bat algorithm and BP algorithm is proposed in this paper and we apply this optimization model in forecasting national economy of Xi’an. The main content of the thesis includes:1.In order to improve the performance of algorithm, a Bat algorithm based on t distribution mutation(referred as TMBA) is present. The TMBA and the Bat algorithm are compared and analyzed by carrying out numerical experiments for 6 test functions. The simulation results show that the TMBA is better than the Bat algorithm.2. The bat algorithm based on t distribution mutation be used for adjusting the structure and parameters of neural network. So a cooperative neural network learning method based on bat algorithm and BP algorithm(referred as TMBA-BP) is proposed. The TMBA-BP and the BP are compared by carrying out simulation experiments for 6 test functions. Experimental results show that the cooperative algorithm proposed can obtain a neural network which has the powerful function approximation capacity. The properties of TMBA-BP are superior to those of the traditional BP model.3.The optimization model of BP neural network based on improved bat algorithm be used for forecasting national economy of Xi’an, compared with the traditional BP, the results of this optimization model is more accurate.