Prediction of chaotic data sequences with BP tuned by an improved PSO

This BP is the most commonly used artificial neural network, but it suffers from extensive computations, relatively slow convergence speed and other possible weaknesses for complex problems. Genetic Algorithm (GA) has been successfully used to train neural networks, but often with the result of exponential computational complexities and hard implementation. Hence Particle Swarm Optimization (PSO) is used to train BP in the paper. For the purpose of predicting chaotic data sequences, an improved PSO is implemented, in which a chaotic way for changing particle velocity is proposed, i.e., the inertia weight is fixed on a chaotic sequence at the beginning of searching process. The efficiency of BP trained with this improved PSO is compared with those of BP and BP tuned with GA based on the prediction of same chaotic data sequences. Comparison based on the searching precision and convergence speed of each method show that BP tuned with PSO is dominant and effective to predict chaotic data sequences.