Global optimisation, model predictive control and uncertainty quantification methodologies for distributed parameter systems

  • Min Tao

Student thesis: Phd


Large-scale distributed parameter systems cover a wide range of practical applications in industrial engineering, such as chemical tubular reactors, bio-production reactors, combustion processes and microscopic reactions on the surface. Intelligent operations including optimisation and control could improve process performance, satisfying process constraints. However, intelligent computations are challenging for large-scale systems due to the global optima issues, process uncertainty and expensive system evaluations. The research in this thesis presents model reduction based global optimisation, model predictive control and uncertainty quantification methodologies for large-scale distributed parameter systems. Firstly, a double reduction (principal component analysis and artificial neural networks) based global optimisation framework was built, which was then improved through piecewise affine and deep rectifier neural network reformulations. Then a combination of nonlinear model predictive control and polynomial chaos expansion was employed to robustly control distributed parameter systems under parametric uncertainty, where the offline reformulations (proper orthogonal decomposition and recurrent neural networks) based global optimisation method was utilised within recursive control steps. Next, an "equation-free" Monte Carlo uncertainty quantification methodology was proposed for large-scale distributed parameter systems, where the recursive projection method and lifting-restriction operations were performed to accelerate the computations of distributional steady states through a black-box dynamic simulator. Finally, a Bayesian optimisation approach was adopted to globally search optimal solutions for large-scale system under uncertainty, where parametric uncertainty was addressed using our presented uncertainty propagation algorithm. The proposed computational frameworks were validated through several practical chemical and biochemical production cases.
Date of Award1 Aug 2022
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorKonstantinos Theodoropoulos (Supervisor) & Jie Li (Supervisor)

Cite this