Modelling, Identification and Control of a Renewable Hydrogen Production System for Mobility Application

Hydrogen-fueled cars are a promising technology for reducing CO2 emissions in the mobility sector. This thesis develops a stochastic receding horizon controller for a hydrogen refueling station that operates the electrolyzer and compressors such that the usage of renewable energies, in this case photovoltaic (PV) energy, is maximized and the usage of grid power is minimized. For this, a model of the electrolyzer and the compressors is derived and identified from data. Historical hydrogen demand data is fitted to a stochastic model such that samples for a scenario-based stochastic MPC can be generated. The available PV power is predicted by a neural network using weather forecast data. These models and predictions are combined into a mixed-integer linear program, which is solved in real-time every 10 minutes. To allow for better scalability of the problem with respect to the number of scenarios, a heuristics to decouple the scenarios in the optimization problem is introduced. The resulting optimization problem is converging within approximately 10 seconds with a prediction horizon length of 24 hours and 96 timesteps. An evaluation of the stochastic MPC and different deterministic controllers shows that the stochastic MPC performs equally well to its deterministic counterpart as long as the storage tanks are not close to their lower limit. In the case of almost empty storage tanks, the stochastic MPC provides an input sequence which leads to a probabilistic satisfaction of the system constraints. Adjustments to the terminal weights in the MPC are proposed in order to increase the performance of the MPC in such situations