Multistep-Ahead Prediction of Time Series Using Independent Models
Zhenming Yang, Jiguang Yue, Yunshi Xiao, Xiaobao Wang · International Conference on Electric Information and Control Engineering · 2012
Multi step-ahead prediction of time series is widely used in real applications, such as finance, transportation and electrical load. A multi step-ahead independent prediction approach of time series is proposed. The step-by-step recurrent approach and the independent approach are compared, and the influence of accumulative error on the performance of multi step-ahead prediction is analyzed. The recurrent neural network is used to realize the independent prediction approach, and the prediction models of urban rail transit are built, trained and tested using MATLAB. The prediction results show that the error of independent prediction is smaller than that of step-by-step recurrent approach. The advantages and disadvantages of each approach are discussed.