Data-Driven Modeling and Predictive Control of a High-Quality Special Steel Electroslag Remelting Process with Time Delay

The position of the consumable electrode in molten slag is changed to control the remelting current and melting rate. The third-order function is used to model the electroslag remelting (ESR) process under the position control operation mode. The model parameters are estimated by the recursive least-squares algorithm. One-dimensional search strategy is applied to determine the nonlinear multisource time delay. The adaptive forgetting factor is constructed by using the rate of change of estimated parameters to ensure a stable convergence of parameters. A model predictive control algorithm based on state augmentation is designed to control the ESR system with a time delay. The time-delay system is transformed into an equivalent time-delay-free augmented state space model by introducing a historical control input sequence into the augmented state vector. The rolling optimization mechanism is utilized to dynamically compensate for the phase lag caused by multisource time delay. The influence of uncertain factors such as thermal inertia of the slag pool is effectively suppressed by feedback correction. Meanwhile, convergence analysis of the proposed control algorithm is also given. By comparison with existing algorithms, the feasibility and superiority of the proposed algorithm are verified by experiments at the pilot base of our laboratory.