The Study Of Hybrid Prediction Model

As a rapidly developing discipline, early warning systems provides theory ap-proach to explore the dynamic characteristics of high-dimensional system by the finite-dimensional information system. In the early warning systems, trend forecasting model is a relatively new, practical idea; In order to monitor and predict the system, an effec-tive model can be built from the time series database.Effective forecasting techniques is the key to improve the performance of the early warning systems, chapter III analyzes several commonly used forecasting model; gen-erally, the parameter selection problems of prediction model are present, and these parameters directly affect quality of the prediction, chapter IV presents several more popular swarm intelligence algorithm, these methods can optimize the parameters of prediction model, thus improving prediction accuracy. Meanwhile, the combination of swarm intelligence algorithms and prediction model is the starting point for model structure optimization.To further improve the prediction accuracy of the model, this paper proposes a hybrid method; and finding the entry point and structure of hybrid model is the key to hybrid modeling. As support vector regression (SVR) has a strong non-linear modeling capability, ARIMA has a strong linear modeling capabilities, we combine these two models to modeling.In this paper, we discussed the differences of different forecasting model; as each model has its own advantages and disadvantages, we propose a hybrid model to im-prove prediction accuracy. Inspired by that the support vector regression (SVR) model, with theε-insensitive loss function, admits of the residual within the boundary values ofε-tube, we propose a hybrid model that combines both SVR and Auto-regressive integrated moving average (ARIMA) models to take advantage of the unique strength of SVR and ARIMA models in nonlinear and linear modeling. A nonlinear analysis of the time series indicates the convenience of nonlinear modeling, the SVR is applied to capture the nonlinear patterns. ARIMA models have been successfully applied in solv-ing the residuals regression estimation problems. The experimental results demonstrate that the model proposed outperforms the existing neural-network approaches, the tra-ditional ARIMA models and other hybrid models based on the root mean square error and mean absolute percentage error.