Application of the Bat Algorithm to Optimize the BP Neural Network

Abstract

For the standard BP algorithm usually has the limitations of slow convergence and local extreme values, a new method to adjust weights of BP network was proposed based on the bat algorithm of the global optimization ability and the strong convergence. The new algorithm was based on the weight adjustments of error back propagation of BP algorithm and the weight and threshold of BP network modification using the bats position update. The new algorithm can not only use the bat ability of global optimization, but also contain the feature of error back propagation of BP algorithm. The new algorithm was used in simulation test of two typical functions, results of which were analyzed and compared with that of basic BP algorithm and PSO-BP algorithm. Experimental results show that the new algorithm has faster convergence speed and higher convergence accuracy, and improved the learning ability and generalization ability of BP network. The performances of the new algorithm were superior to that of other 2 kinds of BP network algorithm.