Multi-step Ahead Time Series Forecasting Based on the Improved Process Neural Networks

This paper proposes a multi-step ahead time series forecasting based on the improved process neural network. The intelligent algorithm particle swarm optimization (PSO) is used to overcome the potential disadvantages of the neural network, such as slow convergence speed and derivative local minima. Firstly, the theoretical analysis of the PSO is given to optimize the multilayer perceptron (MLP) neural network architecture. Secondly, the theoretical analysis and processing flow about the MLP architecture optimization is given. Thirdly, the performance criteria are applied to verify the performance of the proposed approach. Finally, the experimental evaluation based on a typically chaotic time series with rich spectrum information is utilized to demonstrate that the proposed approach has comparative results and superior on forecasting accuracy comparing to the traditional methods.