Hybrid ensemble learning model for predicting external characteristics of proton exchange membrane fuel cells under various operating conditions

Abstract

An accurate and efficient predictive model for external characteristics of proton exchange membrane fuel cells (PEMFCs) is essential for boosting performance and guiding system-level design. In this study, a dedicated PEMFC test bench was designed and influence mechanisms of intake temperature, pressure and relative humidity on cell performance were decoupled and systematically analyzed. On this basis, a hybrid ensemble learning model was proposed to enhance the precision and efficiency of external characteristic predictions. The results demonstrate that elevated intake temperatures improve cell voltage by accelerating reaction kinetics, and low pressures hinder performance through limited reactant supply, while optimal PEMFC performance is achieved at medium humidity levels. Additionally, voltage sampling errors are found to increase under conditions of high temperature, pressure and humidity, reflecting challenges in water management and gas flow regulation. The hybrid ensemble learning model outperforms standalone models, which achieves minimal mean squared errors (MSEs) of 0.2254 for voltage and 1.48 × 10⁻⁴ for voltage sampling error. Its integration of multiple models enhances predictive accuracy and avoids overfitting, demonstrating superior predictive accuracy and adaptability to complex data. These findings provide a crucial data foundation and robust model support for analyzing influence mechanisms of PEMFC external characteristics and accurately predicting performance.