In this thesis, we address the biopharmaceutical production scheduling problem in which protein products should be produced to maximize the profit, considering the frozen production period and weekend break, facing the uncertainties of non-linear product concentration and decay of resin’s purification capacity. We suggest modeling the corresponding production scheduling problem using an infinite-horizon Markov Decision Process which allows for a near-optimal solution in the case study using an efficient Q learning algorithm of reinforcement learning. A grid search is designed to find the appropriate hyperparameter combination of the algorithm. Results show a slight weekend effect: considering weekend breaks does not have a significant impact on overall profits, compared to not considering a weekend break in production scheduling. In addition, we observe a trend that the longer the frozen production period, the lower the overall profit. The main reason is that a longer duration of frozen horizon leads to more frequent resin exchange operations, which is the solution to avoid the risks of a potential reduction in batch product yield due to reduced purification capacity. We also find that there is no fixed pattern emerged in the production schedule results.     «

In this thesis, we address the biopharmaceutical production scheduling problem in which protein products should be produced to maximize the profit, considering the frozen production period and weekend break, facing the uncertainties of non-linear product concentration and decay of resin’s purification capacity. We suggest modeling the corresponding production scheduling problem using an infinite-horizon Markov Decision Process which allows for a near-optimal solution in the case study using an e...     »