Low cost, high performance fuel cell energy conditioning system controlled by neural network

Abstract

Fuel cells are an important option for the generation of renewable, efficient and environmentally friendly electricity. Although there are commercial applications in the industrial, residential and automotive sectors, it is not yet a mature technology and requires much research, particularly to reduce its costs to a level competitive with other technologies. This research is currently focused not only on the structure of the cell but also on the additional elements and subsystems required for its implementation as an energy solution. In this article, we propose an electrical energy conditioning scheme for the Formic acid fuel cell (direct formic acid fuel cell or DFAFC). This fuel cell was selected for its high performance, and low cost in low and medium power applications. The proposed system consists of a direct current-direct current (DC-DC) regulator supported by a power converter controlled by a Cortex-M3 ARM processor. This CPU is used to propagate a static neural network trained with the non-linear dynamics of the power converter. The power circuit is modeled and simulated to produce the training parameters. The neural network is trained externally and runs off-line on the processor. The results show not only the regulation capacity of the control scheme but also its response speed to sudden changes in the load.