The Multi-Rule & Real-Time Training Neural Network Model for Time Series Forecasting Problem
Xizheng Zhang, Lining Xing · 2006
In view of the limitation of existing neural network model in solving time series forecasting problem, put forward a new multi-rule & real-time training neural network (MRRTTNN) model. The characteristics of this proposed model including (1) miniaturize the forecasting network, (2) train the network in real-time way, (3) adopt the average of abundant forecasting and (4) add some rules to assistant forecasting. Relative to the traditional neural network model, this model focus on dynamic training and dynamic forecasting, increase three rules (rule of dealing with abnormity, rule of retraining and rule of adopting the average) to assistant forecasting. Numerical example suggests the correctness and feasibility of this model. The contradistinctive result of this model and other five models indicates the validity and superiority of this model