This paper discusses the novel approach of introducing artificial neural networks into detailed simulations of complex biochemical processes in order that available deterministic mathematical modelling techniques can be applied with enhanced benefit. The mechanistic part of the hybrid model is composed of 44 variables and 145 parameters. The neural networks were designed to adjust the key 15 parameters of this model with a reinforcement learning scheme since the desired output vector of the neural network, or the most appropriate model parameters, is unknown. The combination of the first principle model and artificial neural networks has yielded better model predictions for substrate consumption, toxic by-product accumulation, cell growth and cell composition, and metabolic product formation than that using either of the approaches independently.