Fuel Cell Stack Power Prediction Model Using Gaussian Process Regression Mode

A fuel cell stack power prediction model that takes into consideration the various stack control parameters is important in the optimization design of the controls for each item of auxiliary equipment in a system equivalent to that of an actual vehicle. However, creating a model for quantitative prediction of stack power requires large amounts of data concerning the materials and structure inside the fuel cell.

Moreover, since the internal phenomena are complex, large-scale modeling is necessary. For this research, a design of experiment method known as the space filling technique was used to acquire data efficiently.

With the acquired data as a basis, the use of Gaussian process regression made it possible to create a model capable of predicting stack performance as well as the temperature and pressure in the various parts of the stack in a short computation time.

It was also made clear that this model could be used to calculate operating conditions that would maximize stack power, and verification by testing showed that it would be possible to obtain a power prediction model that could be used to investigate stack performance from a limited amount of test.

ETAS ASCMO was used to optimize the fuel cell stack performance considering all control parameters and using INCA-FLOW the test process was automated. As a result, our customer Honda achieved a 6% increase of gross power.