Modelling plays an increasingly important role in chemical and bioprocesses nowadays and is widely used for process simulation, optimisation and real-time control. Especially for metabolic reactions with complex underlying reaction mechanisms, modelling for process analysis, prediction and control is a very cost-effective technique. In this MPhil project, a temperature-dependent kinetic model to simulate biomass growth, substrate consumption and the production of GLA by Cunninghamella echinulata was first proposed. The model was verified to be of high accuracy using data from a 1L bioreactor. Model aided upscaling to a 5L bioreactor with a two-stage temperature-shift strategy showed a 69.6% increment of GLA production and was verified experimentally. Then, hybrid modelling which is a state-of-the-art modelling technique and combine machine learning techniques and traditional kinetic models, was used to simulate and predict the performance of the GLA fermentation experiment by Cunninghamella echinulata. In addition, the hybrid models incorporated different amounts of kinetic information from a pre-existing complex kinetic model, representing different level of hybrid model âgreynessâ was investigated for bioprocess predictive modelling. The results show that incorporating more specific kinetic information increased the risk of incorporating incorrect inductive bias that hindered rather than enhanced hybrid model performance. Nonetheless, the hybrid models demonstrated much improved predictive confidence with similar predictive accuracy to the original kinetic model.
|Date of Award
|1 Aug 2023
- The University of Manchester
|Robin Smith (Supervisor) & Dongda Zhang (Supervisor)