A least square support vector machine prediction algorithm for chaotic time series based on the iterative error correction

This paper analyzes the error characteristic of traditional support vector machine prediction algorithm, where the error series are smooth and regular. This is because a single prediction model is incapable of fitting chaotic system mapping function and omitting some deterministic component. On this basis, a prediction algorithm that consists of an iterative error correction and a least square support vector machine (LSSVM) is proposed. The algorithm creats multiple predictive models via the method of iterative error correction to approximate the chaotic system mapping function and obtain significant improvements of predictive performance. In addition, the optimal parameters of the prediction model are automatically obtained from the pattern search algorithm which is simple and effective. Experiment conducted on Lorenz time series and MackeyGlass time series indicates that the proposed algorithm has a much better performance than that recorded in the literature.