Study of the performance of the multi-level iteration scheme for dynamic online optimization for a fed-batch reactor example

摘要

Nonlinear Model Predictive Control (NMPC) is an advanced control technique which finds increasing interest in industry. One obstacle to its more widespread use is the computational effort due to the resulting nonlinear dynamic programming problems. Efficient online optimization methods are required to overcome this problem, in particular for processes that require fast sampling times. A promising approach to online optimization is the Multi-Level Iteration (MLI) scheme that was proposed in [1]. The MLI scheme is based on Sequential Quadratic Programming (SQP) and consists of four solution modes, which differ in the performance and computation speed due to the amount of information that is used when solving the quadratic programming (QP) subproblems. It has been successfully tested on theoretical case studies. In this work, the MLI scheme is for the first time investigated experimentally for solving the optimal statefeedback control of a nonlinear fed-batch process where the thermal system is real hardware and the chemistry is considered by inserting the resulting heat of reaction via a heating device. The performance of the MLI scheme is studied for different combinations of modes and different frequencies of using each of them, and the resulting control performance is compared.