| 초록 |
The model-plant mismatch learning-based offset-free model predictive control (MPC) is proposed to overcome the limitation of model-based and data-driven control methods by combining both strategies based on the offset-free control framework. In this study, the intrinsic model-plant mismatch map is derived via a data-driven machine learning approach and then utilized in the model-based controller. Specifically, first, an artificial neural network is trained to approximate the model-plant mismatch on the steady-state manifold in the controlled variable space. Then, a revised form of disturbance estimator is constructed to compute the supplementary disturbance variable while considering the influence of the mismatch value from the learned mismatch map. Subsequently, the mismatch information and the supplementary disturbance are applied to the target problem and the finite-horizon optimal control problem. This allows the controller to utilize both the data-driven mismatch map via machine learning and the stabilizing property of the disturbance estimator. The closed-loop simulation result shows that the developed scheme can efficiently improve the reference tracking performance. |
