"Data based modeling and optimal control of a PEM fuel cell system"
Data based modeling and optimal control of a PEM fuel cell system
Sprache des Titels:
This diploma thesis includes the data based modeling of a Proton Exchange Membrane (PEM) fuel cell system (FCS) and an optimal control design for this FCS.
For data based modeling the necessary data are taken from an existing FCS which is operating in an experimental enviroment at the Fronius GmbH. The considered FCS consists of a PEM stack and the for operating necessary facilities such as compressor, cooling circuit and an electrical load. The peak power of the stack is 5 kW and its main purpose is for a mobile use.
Throughout this thesis mainly the air path system and the cooling circuit are considered.
For both systems it is shown how to identify a data based model which is suitable for control tasks. The data based modeling includes preprocessing, system structure identification and parameter estimation or identification. Although for structure identification different approaches such as for NARX identification are applied, it turns out that for this present application a self made definition of the system structure considering physical aspects gives the best result.
After these steps the final system identification is performed by applying the well known prediction error methode (pem). Because of the different time scales two different models are identified, one for the air path and one for the cooling circuit. These models are then applied to design a controller.
Therefore for the air path system the control problem is rewritten in a H-inf tracking set up and solved by using standard algorithms.
For the cooling system, since of its huge delay time, a Smith Predictor combination with a PID controller is performed. Both control results show a well performance in simulation.
Finally it can be summarized that not only the performance of the controllers for the air path system and for the cooling system are important to increase efficiency but mainly the set points of the operating points are important. To this end a stationary optimization is performed.