Adaptive modeling of rolling force for hot rolled plate based on industrial data

Abstract

To reduce the prediction error of traditional Sims rolling force model (abbreviated as the Sims model), a new prediction model of hot rolling force is established by modifying the Sims model with the adaptive exponential smoothing method using industrial-measured data. Firstly, statistical analysis of industrial-measured rolling force is carried out, and the discrete distribution characteristic of the data is revealed. Then, an adaptive correction factor is introduced to the Sims model. By using the single-parameter exponential smoothing method, the average error of the predicted rolling force is reduced. On this basis, to adapt the model to the discrete data and prevent great errors, a trend adjustment parameter is introduced into the adaptive correction term so that the correction term can be updated accurately in the adaptive process. Finally, a new rolling force model is obtained. The new model achieves a prediction error of 3.75 % when validated with Q345 steel measured data, which is lower than that by the Sims model of 12.68 %. Meanwhile, to further reduce the energy consumption, the multi-objective particle swarm optimization algorithm is used to optimize the rolling schedule, which can reduce the rolling power by 20.07 %–30.04 %, and the reasonable assignment of rolling force during rough rolling stage is realized. The present research can provide scientific guidance for both the construction of a high-precision rolling force model and the optimization of hot rolled plate rolling schedule.