Using GA-BP Neural Network for Sticking Breakout Prediction in Continuous Slab Casting

Breakout is a hazardous accident in continuous casting process. Losses caused by a typical breakout accident can be as high as hundreds of thousands of dollars. Breakout prediction technique is playing an important role in reducing the occurrence of breakout events. Back propagation (BP) neural network based forecast method as an important breakout prediction technique which can be applied in continuous casting. Due to the error function of BP neural network not being a strictly convex function, when training BP neural network it is easy to fall into a local extreme point. In this paper, single thermocouple time sequence network and “T” shaped four thermocouples space network are constructed for sticking breakout prediction. A genetic algorithm (GA) is used to determine the initial values of network weights to make BP neural network converge to global optimum more quickly and not to plunge into a local extreme. Compared to previous breakout prediction methods our results show it is more accurate.