Abstract:

A multi-steps forecast model possessing high precision based on procedure neural network (PNN) is established. Aiming at the complexity of training PNN, a new algorithm based on combining orthogonal function basis expansion and vector distance based immune algorithm (VD-IA) is proposed. The mathematic model of PNN that is expressed based on orthogonal trigonometric function basis is used to deduce the optimization model suitable to the VD-IA. An adaptive strategy is designed to obtain quick convergence process. The validity of the proposed method is vertified by Mackey-Glass chaotic sequence and is compared with both BP algorithm and improved particle swarm optimization (IPSO) algorithm. Simulation results show that the outstanding results can be obtained by using VD-IA, and the generalization performance of the IA-PNN is also outstanding.