Hybrid time series interval prediction by granular neural network and ARIMA

Granular Computing has been successfully applied in many fields to assist modeling while uncertainty exists due to its intrinsic flexibility and comprehensive evaluation on different formats of information granules. The study of residual modeling calls for some flexible tools like Granular Computing techniques to capture the high uncertainty intra the residual. In this study, we devote ourselves to designing a hybrid time series prediction model by developing linear series model and nonlinear (residual) series model separately. After that, a linear information granule aggregation strategy is adopted generate the final predicted result. We focus on the developing a granular model (neural network) to capture the pattern among the highly nonlinear residual and providing an interval residual with upper bound and lower bound which covers as many training data (real residual) as possible and keeps specific enough simultaneously. To verify the effectiveness of our method, a series of experimental studies are executed on several benchmark time series and compared with the ARIMA (Autoregressive Integrated Moving Average) model. Our predicted intervals show a more compact results under the same coverage.