Research On Billet Temperature Modeling Method Of Reheating Furnace Based On Neural Network

As a vital factor, the temperature of the outlet strip affects the quality and flatness of the strip derectly. It can assure the rolling quality and reduce energy consuming in heating furnace if the billet’s temperature is control reasonably. What’s more, the shortage of energy sources is seriours nowadays. Therefore, it is significant to establish a valid model of the furnace, calculateing the outlet strip temperature.The working pinciple of heating furnace and intelligent algorithms were introduced in this paper, which is under a background of researching the heating furnace. As for the problems of the difficulty in computing, the numerous parameter and hard to determine in traditional modeling methods, RBF neural netwoek intelligent method for establing forecast model is applied in this paper. K-means clustering (K-means), differential evolution (DE) and orthogonal least squares (OLS) method were used for optimizing the model. The simulation proved the validity of the four forecasting models, but certain problems still existed. Therefore, a new steel temperature prediction model based on combinating both differential evolutions orthogonal least squares algorithm (DEOLS) and RBF Neural Networks is proposed in the paper. DEOLS is the algorithm that integrates OLS and DE naturaly. DE assists OLS to code for hidden node centers and width, which set decoded output array as an alternative set; DE evaluates individual advantages and disadvantages of population through OLS. The algorithm can be more reasonable to determine the number of hidden nodes RBF model, the center and width. It reduced the number of iterations significantly.The simulation shows that the DE-RBF forecast model poposed in the paper effectively overcomes the shortcomings of RBF network model on the aspect of big error; it also effectively overcomes the shortcomings such as in instability, large structure and difficulty in calculating of K-means-RBF, OLS-RBF, DE-RBF network model respectively. It proves that the differential evolution-orthogonal least squares algorithm has a better efficiency and higher quality than the single algorithm or other heuristic search methods in solving some complex problems. What’s more, it is a well trial for the researching on a bellet temperature mathematical model in the future.