Nonlinear hybrid control model for accurate steady-state predictions

Abstract

Since the industrial sector accounts for a considerable share of total energy consumption, industrial plants must be improved in terms of their energy efficiency. Improved process control can be seen as one approach towards achieving this goal. These controllers are based on a control model and thus are able to optimize future plant performance. Mostly linear control models are used, though, especially for highly nonlinear processes, nonlinear control models are advantageous. While physical modeling of those can be challenging, data-driven control models can also be used. However, it is shown that even these lead to slightly inaccurate predictions especially for steady-state operation. Therefore, a hybrid model is developed starting from a data-driven control model before a physical model component is introduced. As a result, the hybrid model is able to reproduce the correct steady-state plant behavior, thus improving the overall prediction quality of the control model.