Abstract
In order to accelerate the commercialization of solid oxide fuel cells, optimal process parameters for reliable and efficient electricity generation are of the highest interest. To reduce the number of timely and monetarily expensive experiments used to find suitable operating parameters, while also allowing for only small sacrifices in accuracy, an artificial neural network (ANN) is used in combination with algorithm-based optimization in this study. An ANN was trained with data from a complex multi-physics model and coupled with a genetic algorithm (GA) to find the maximum power output within a range of operational parameters. Instead of time-consuming, manual fine-tuning of the ANN's model architecture and hyperparameters (HP), a Bayesian HP-tuning algorithm is used in this work. To avoid over-fitting and to ensure a high model consistency, nested k-fold cross validation is implemented. Very low error values of the ANN are achieved both in the k-fold cross-validation and in an additionally performed validation by means of experimental data and a computational fluid dynamics simulation. Compared to a multi-physics model used to generate training data, the ANN achieved an increased prediction speed of more than three orders of magnitude, while only minimal decreases to prediction accuracy. Optimization with the GA produced consistent results close to the global optimum and also provided good alternative solutions with significantly different gas compositions at high power. The validity of the solutions found with the GA was underpinned with the help of a sensitivity analysis, which was carried out for the most promising SOFC operating case.
Original language | English |
---|---|
Article number | 117263 |
Number of pages | 13 |
Journal | Energy Conversion and Management |
Volume | 291 |
Early online date | 17 Jun 2023 |
DOIs | |
Publication status | Published - 1 Sept 2023 |
Keywords
- Artificial neural network
- Genetic algorithm
- Optimization
- Solid oxide fuel cell
- Time-efficient prediction
ASJC Scopus subject areas
- Renewable Energy, Sustainability and the Environment
- Nuclear Energy and Engineering
- Fuel Technology
- Energy Engineering and Power Technology