Abstract: As the stress state is complex in tandem cold rolling, traditional mathematic model of rolling force cannot meet the requirement of dimensional precision, by analyzing influential factors, the parameters of input layer and the number of hidden layer nodes are selected before the radial basis function(RBF) neural network model for stress state coefficient is established. After that, the stress state coefficient model is combined with the traditional mathematic rolling force model, as a result, a rolling force revised model is obtained. There are shortcomings in adaptive genetic algorithm (SAGA) proposed by Srinvas such as poor local search ability ,an improved adaptive crossover and mutation strategy is designed to make the load equal in each housing. Comparison to the schedule is offered, experimental results on 1370 five tandem cold rolling with two typical steel grades demonstrate that the improved adaptive genetic algorithm (IAGA) possesses faster speed and higher reliability than adaptive genetic algorithm proposed by Srinvas. The standard deviations of load coefficient of front four housings reduce to 0.010 8 and 0.009 0 respectively.

Key words: Adaptive geneic algorithm, Radial basis function neural networks, Schedule optimization, Tandem cold rolling