Multiple time series autoregressive method based on support vector regression

Energy consumption time series involves a variety of energy and the relationship between different energy is complicated.Most existing consumption methods make prediction through multiple independent single time series respectively,which ignores dependencies between multiple time series.In order to take full advantage of the association between multiple time series and improve prediction accuracy,the vector-valued autoregressive method and multi-task autoregressive method based on Support Vector Regression(SVR) machines were proposed for multiple time series forecast according to vector-valued function learning and multi-task learning theory.The experimental results with energy consumption of coking process verify that multiple time series autoregressive models based on the proposed methods show better prediction performance.