Personal description

This is Zhengang Zhong’s personal webpage. My given name is pronounced like /ZHən gang/ (ZH in decision).

About me

Welcome! I am a PhD student at Imperial College London in the Optimisation and Machine Learning for Process Engineering lab. I conduct my reserach on the data-driven optimal control under the supervision of Antonio del Rio Chanona and Panagiotis Petsagkourakis. I obtained my Diplom degree in Mechatronics from Technische Universität Dresden and my B. Eng. from Zhejiang University of Technology. I am a big fan of control theory and German beer.

About my research

My PhD research focuses on the data-driven model predictive control (MPC). MPC is used in many engineering applications including bio-production processes, food processing and CO2 capture and storage. In the scenario of process engineering, MPC suffers from plant-model mismatch and disturbances. This research aims to investigate data-driven MPC for chemical processes to address these problems. This research project consists primarily of two interdependent subtasks: 1. construct data-driven multi-level modelling frameworks for chemical processes; 2. apply the frameworks into a variety of applications (e.g. yeast, algae and cyanobacterial production).

To diminish the risks arising from process disturbances and model mismatch, this research will explore cutting-edge data-driven approaches. Most chemical processes are stochastic and operate dynamically; therefore, optimisation strategy combining MPC and reinforcement learning will be proposed. The project will mathematically analyze stability and robustness of the optimisation algorithm and demonstrate control performance through simulation. Since data-driven optimisation algorithms are data-hungry, designing state observers and dealing with ill-posed problems will be instrumental to the research. An optimal control framework - comprising measurement noise detection, optimisation strategy computation and system dynamics simulation - will be an important intermediate product of this research.

This research project will apply the frameworks into real-world applications. To achieve better control performance of real systems and reduce design time and cost, optimal cyber-physical co-design of controlled systems will be explored. This project is also aimed at proposing more efficient real-time nonlinear programming approaches to address a large class of nonlinear optimisation problems. The applied control approaches will be compared with other existing methods (e.g. PID or MPC) with regard to control benchmarks such as convergence time and steady state error.