An improved PSO-BP neural network and its application to earthquake prediction

This paper presents a way of combining BP (Back Propagation) neural network and an improved PSO (Particle Swarm Optimization) algorithm to predict the earthquake magnitude. It is known that the BP neural network and the normal PSO-BP neural network have some defeats, such as the slow convergence rate, easily falling into local minimum values. For improving the properties of PSO, some proposed the linear decreasing inertia weight strategy. Furthermore, this paper uses a nonlinear decreasing inertia weight in PSO to get a faster training speed and better optimal solutions. Compared with the linear decreasing strategy, the inertia weight in our nonlinear method has a faster declining speed in the early iteration, which can enhance the searching precision. In the late iteration, the inertia weight has a slower declining speed to avoid trapping in local minimum value. Then we apply the improved PSO to optimize the parameters of BP neural network. In the end, the improved PSO-BP neural network is applied to earthquake prediction. The simulation results show that the proposed improved PSO-BP neural network has faster convergence rate and better predictive effect than the BP neural network and the normal PSO-BP neural network.