Guiding the optimization of membraneless microfluidic fuel cells via explainable artificial intelligence: Comparative analyses of multiple machine learning models and investigation of key operating parameters

Membraneless microfluidic fuel cells (MMFCs) offer great potential for clean energy production, but their expense and tedious optimization process have limited their wider use. Machine learning (ML) algorithms have shown promise in improving the optimization of MMFCs, but the “black box” nature of these models has caused uncertainty and limited their adoption. To address this, we conducted the first study that implements the explainable artificial intelligence (XAI) approach to gain in-depth insights into multiple ML optimization models' predictions and the impacts of operating parameters on MMFCs' performance. Among 27 investigated models that are generated based on nine ML and three bio-inspired evolutionary algorithms, the combination of a decision tree and the particle swarm optimization algorithm is the most effective model that satisfied all criteria of high time efficiency (execution time = 3.194 s), accuracy (squared correlation coefficient = 0.999), and optimum power density identification (0.278 mWcm⁻²). The ML-optimized power density is 239.024% increased, which is 3.4 times higher than the average power density obtained without optimization. We further utilized the XAI model to comprehensively examine the impacts of operating parameters, providing valuable decision aids for the researchers, technicians, and engineers in manufacturing optimal power MMFCs at a significantly lower cost and faster rate.