A Data-Driven Multiobjective Dynamic Robust Modeling and Operation Optimization for Continuous Annealing Production Process

Abstract

There are many dynamic disturbances during the continuous annealing production line (CAPL) in iron and steel enterprise. Traditional robust operation optimization considers only the maximum disturbance range in previous production and overrides the dynamic changes of these disturbances, which often results in high production cost and low product quality. Therefore, this paper proposes a novel multiobjective dynamic robust optimization (MODRO) modeling method by further taking into account the dynamic changes of these disturbances and adopting a time series prediction model based on a least square support vector regression (LSSVR) to predict the range of disturbances in next time slot. The main feature of the model is that the robustness can be dynamically adjusted according to the disturbance range predicted by the LSSVR. To solve this model, an improved NSGA-II algorithm is developed based on a new crowding metric. Numerical results based on actual production process data illustrate that the proposed MODRO modeling method is obviously superior to traditional static robust operation optimization, and that it can significantly improve the strip quality and the capacity utilization of the CAPL, and reduce the total energy consumption.