Predictive Modeling of Li-Air Batteries Using Artificial Neural Network: A Comparative Study of Cathode Morphology

Abstract

The artificial neural network (ANN) modeling is used to analyze the impact of two different cathode morphologies urchins (α-MnO₂ ) and flower (δ-MnO₂ ), on the charge/discharge voltage in lithium air batteries (LABs). Previous research has focused on ANN models for traditional lithium-ion batteries (LIBs) without accounting for varied cathode morphologies in LABs. This research presents an ANN modeling technique to predict the charge/discharge voltage LAB using manganese oxide as cathode materials with two distinct morphologies. For modeling Specific capacity use as the input variable, to perform a comprehensive analysis to validate charge/discharge voltages. This study explores multiple ANN configurations with varying neuron counts, identifying the optimal architecture (10 neurons in hidden layers) that balances prediction accuracy and efficiency. This systematic exploration provides insights into ANN tuning for LABs, which is a topic with limited coverage in existing literature. The ANN predicted results closely matched with the reported experimental work with the coefficient of determination R² = 0.9998 for almost all models. The models performance was assessed by various error metrics mean absolute deviation (MAD), root mean square error (RMSE) and average absolute relative error (AARE). This study provides empirical validation of the model's robustness. The study highlights the applicability of ANN in capturing complex LAB performance metrics, such as the non-linear behaviors due to morphological differences.