Optimal flexible operation of dynamic processes

  • Optimale flexible Betriebsführung dynamischer Prozesse

Caspari, Adrian; Mitsos, Alexander (Thesis advisor); Biegler, Lorenz T. (Thesis advisor)

Aachen : Aachener Verfahrenstechnik (2021)
Book, Dissertation / PhD Thesis

In: Aachener Verfahrenstechnik Series, Process Systems Engineering 14
Page(s)/Article-Nr.: 1 Online-Ressource (XVII, 218 Seiten) : Illustrationen, Diagramme

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2021


Flexible operation of energy-intensive processes is economically and ecologically promising due to the existence of fluctuating electricity prices, as induced by the integration of renewable sources in the energy grids. Grid stability, price fluctuations, and the desire for the best possible process behavior motivate an optimal flexible operation paradigm to realize optimal demand side management (DSM). In case characteristic process dynamics and electricity price fluctuations lie in similar time horizons, the explicit consideration of the process dynamics for optimal flexible operation is necessary. This holds, e.g., for air separation units (ASUs) and the German Day-Ahead auction and inspires considering these as applications. In this dissertation, we, therefore, address several aspects of the optimal flexible operation of dynamic processes using model- and optimization-based methods. In the beginning, we propose a multi-product ASU process design that is tailored to DSM, as the process power demand can be varied by more than 80 % during operation. Subsequently, we present an approach for the solution of dynamic optimization problems with complementarity constraints. Using smoothed nonlinear complementarity problem functions allows solving these optimization problems with standard direct shooting methods. The proposed approach can be used to formulate and optimize dynamic systems with nonsmooth behavior. In particular, we use the method to optimize the start-up procedure of a single-product ASU.While the first chapters focus on offline optimization and process design, we turn to optimal process operation by model predictive control in the remaining chapter. We compare two paradigms for the optimal flexible operation of dynamic processes, i.e., top-down vs. bottom-up, showing that (economic) nonlinear model predictive control ((e)NMPC), referred to as bottom-up, is computationally more demanding but superior in terms of performance over integrated scheduling and control based on scale-bridging models, referred to as top-down. As NMPC comes with computational burdens that often prevent its real-time capability, we present two strategies to reduce the computational burdens. As a first possibility to reduce the CPU times for dynamic optimization, we apply a suboptimal fast-update approach for eNMPC to a single-product ASU. We then illustrate the flexibility of the complex, previously presented flexible ASU by applying a similar fast-updated based eNMPC and compare the performance of eNMPC with optimal quasi-stationary scheduling. The first outperforms the latter. The second strategy is the application of dynamic model reduction techniques. We propose a reduced distillation column model based on wave propagation that clearly reduces the computation burden for optimization and NMPC. Using a reduced dynamic controller model as well as almost any application of NMPC in reality induces a plant-model mismatch. This leads to an undesired offset in NMPC, i.e., controlled variables (CVs) permanently differ from their desired setpoints. This offset can be prevented by offset-free NMPC, where a nominal model is extended by a disturbance model. We develop the first general approach for the generation of disturbance models in offset-free NMPC based on semi-infinite programming. Further, we extend the theory about offset-free NMPC by allowing for more measurements than CVs and for non-measured CVs, cases that often appear in chemical engineering applications.