A distributed support vector machines architecture for chaotic time series prediction

Chaos limits predictability so that the prediction of chaotic time series is very difficult. Originated from the idea of combining several models to improve prediction accuracy and robustness, a new approach is presented to model and predict chaotic time series based on a distributed support vector machines in the embedding phase space. A three-stage architecture of the distributed support vector machines is proposed to improve its prediction accuracy and generalization performance for chaotic time series. In the first stage, Fuzzy C-means clustering algorithm is adopted to partition the input dataset into several subsets. Then, in the second stage, all the submodels are constructed by least squares support vector machines that best fit partitioned subsets, respectively, with Gaussian radial basis function kernel and the optimal free parameters. A fuzzy synthesis algorithm is used in the third stage to combine the outputs of submodels to obtain the final output, in which the degrees of memberships are generated by the relationship between a new input sample data and each subset center. All the models are evaluated by coal mine gas concentration in the experiment. The simulation shows that the distributed support vector machines achieves significant improvement in the generalization performance and the storage consumption in comparison with the single support vector machine model.