A prediction model based on adaptive selective scenario clustering and error correction

To improve the prediction accuracy of multivariate time series, this paper presents a new prediction model, which contains an adaptive scenario clustering (ASC) - adaptive selective prediction (ASP) method and MPCSO-SVM based error correction method. Firstly, ASC clusters the factors of multivariate time series to obtain the multiple scenarios, and classify the new multivariate time series samples to different scenarios. Secondly, ASP chooses the most suitable algorithm to predict for each determined scenario by comparing the root mean square relative errors of several potential prediction algorithms: Radical Basis Function Neural Network (RBFNN) and Least Squares Support Vector Machine (LSSVM). Thirdly, the Modified Parallel Cat Swarm Optimization (MPCSO) is used to optimize SVM parameters, to correct the prediction error. The simulation results show that the results of prediction with ASC are more accurate than the prediction without ASC method, ASP after ASC improves the prediction accuracy further, and the MPCSO-SVM error correction method improves the prediction accuracy on the basis of ASC-ASP.