A new Integrated Model and its Application to Soft-sensing of the Flue Temperature in Coke Oven

Based on the features of coke oven flue temperature, a new integrated model combining temporal difference method (TD), linear regress(LR) and elman neural network (ENN) is proposed. Firstly, LR models with one variable, two variables and twelve variables are built base on the relationship between the flue temperature and top of regenerators´ temperature, and rationally integrated by elman neural network (LR-ENN). Comparing to the unique LR models, the integrated model shows the good performance. Then modified elman neural network model based on the temporal difference method is used(TD-ENN). Through this model, the error of the LR-ENN is predicted multi-step ahead. At last, the flue temperature is get through the expert coordinator which is used to coordinate the outputs of LR-ENN and TD-ENN. The actual results confirm the integrated model´s validity.