Abstract Decision support tools are essential for planning end-of-life pathways of electric vehicle (EV) batteries, including recycling, repurposing, and reconditioning. As large volumes of EV batteries reach end-of-life in the coming decades, estimating material flows under different scenarios becomes increasingly important. Detailed simulation models, such as those based on discrete event and agent-based modeling, can capture the complex dynamics of battery life cycles. However, their high computational cost limits their applicability in decision-making contexts that require extensive scenario analysis. To overcome this challenge, machine learning-based surrogate models are evaluated for approximating the outputs of a simulation model that estimates future end-of-life quantities of EV batteries. The simulation model uses static input configurations and produces time-dependent outputs, for which no historical time series exist. Two surrogate modeling strategies are compared: an all-at-once approach that predicts all time steps simultaneously, and a recursive approach that predicts them sequentially. Several machine learning algorithms are evaluated, including neural networks, Gaussian process regression, random forests, and XGBoost. Results show that neither approach consistently outperforms the other, and model performance varies strongly across key performance indicators. While surrogate models can substantially accelerate analysis, predictive accuracy is limited for outputs with sparse or highly variable data. The results underline the importance of selecting surrogate modeling approaches based on the specific characteristics of the output indicators and the analytical needs of the decision context.
Huster et al. (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: