Abstract: The production mode of cold rolled strip is changing to multivarieties, small batch and low inventory, which directly results in the need to change the varieties specifications more frequently during rolling. Because changes in product specifications and rolling process status are large, the traditional mechanism model and conventional self-learning and adaptive method are difficult to ensure the setting accuracy of the first volume product. In order to improve the prediction accuracy of rolling force in the process of variable specification, this paper puts forward a composite model based on mechanism model and mass history data of rolling process. The new rolling force model, based on the theoretical model of rolling force, is corrected by BP neural network optimized by genetic algorithm (GA). The new method for prediction of the rolling force makes the average relative error of the first specification of rolled steel rolling force prediction be controlled within 5.5%. The setting accuracy is much higher than that of the conventional mechanism model.

Key words: customized production, rolling force prediction, genetic algorithm, BP neural network