Comparison and general law research of multiple machine-learning models for proton exchange membrane electrolytic cell parameters prediction

This paper presents a simulation-based framework for predicting the performance of proton exchange membrane electrolytic cells (PEMEC). Machine learning techniques are employed to conduct predictive modeling and comparative analysis, with the aim of identifying the optimal machine learning model for evaluating PEMEC parameters. By establishing a three-dimensional PEMEC physical model, the study analyzes the effects of voltage, inlet water flow rate and temperature, membrane thickness, anode gas diffusion layer porosity and thickness, and anode catalyst layer conductivity on the three performance evaluation indicators of current density, hydrogen mole fraction, and temperature, and a dataset of 2061 parameter-value pairs is generated. Correlation analysis of the evaluation index is carried out and eight different machine learning techniques are adopted for prediction, including Elman neural network, back propagation neural network, long short-term memory, random forest (RF), support vector machine, bidirectional long short-term memory, Firefly algorithm optimized Elman neural network, and genetic algorithm optimized back propagation neural network (GA-BP). The obtained results show that voltage, membrane thickness and inlet water temperature mainly affect three performance evaluation indicators. RF algorithm (an ensemble learning framework) and GA-BP (a model integrating global search and parallel computation mechanisms) are identified as the optimal prediction models, with determination coefficients of 0.995, 0.992, 0.992 and 0.996, 0.997, 0.993 for the three indicators, respectively. These findings highlight the superiority of ensemble learning frameworks and optimization-based models with global search capabilities in predicting PEMEC parameters. This study provides a reference for selecting the best machine learning model for predicting PEMEC parameters.