Neural network controller for regulation of a water cooled fuel cell stack

In this thesis, a neural network is designed to control the stack temperature of a fuel cell. A 5kW proton exchange membrane fuel cell is studied with varying current demand resulting in temperature changes in the stack. This controller is initially trained to recognize the pattern of dynamic power demand from a consumption source and the amount of water required for the corresponding demand. The power demand leads to current production by the fuel cell, resulting in temperature increase in the fuel cell stack. Cooling water is used as a coolant to maintain this temperature change. The flow of cooling water is controlled by the controller to maintain temperature at set point. The results show that the neural network controller does an adequate job in maintaining the stack temperature within 4K of the set point temperature. The results are compared to those of a traditional PI controller and a time varying PI controller and it is observed that the neural network provides better results than both the PI controllers.