A hybrid heuristic searching approach for dynamic system modeling is presented. The paper suggests that the model consists of two function parts---GAs and heuristic random searching algorithm (HRSA). GA is one of the adaptive search algorithms which are able to find global solutions or regions in optimal problem. This character is helpful for reducing the searching range in many optimal problems. Based on this foundation, the solutions within these separate regions will be located further by HRSA. Heuristic information is used to form the next possible searching directions in virtue of the gradient concepts. It reduces the computing time of modeling and speed up the identification of the nonlinear dynamic system. Sereral functions are used to test. The results and analysis are discussed. It shows the ability of model in the dynamic system modeling with the features of simplicity and flexibility.

This paper presents a hybrid learning approach for dynamic system modelling and prediction using neural networks. The model learning is divided into two parts. One is to select the global region and the other is to find the goal value. A heuristic learning algorithm (HLA) is discussed, which is effective in the real-time dynamic modelling and control. The hybrid model is applied to the on-line prediction of the rolling strip in the hot skip-pass process. The control system is introduced and the result is discussed.

No abstract