Smartphone battery drain is governed by coupled effects of workload, electrochemical aging, and thermal feedback. Nonlinear behaviors such as voltage collapse remain challenging for traditional models. An electrochemical-informed equivalent-circuit and lumped-thermal continuous-time framework is developed by integrating an equivalent-circuit voltage model with lumped thermal dynamics, aging-aware resistance and capacity evolution, driven by a modular decomposition of smartphone power into CPU load, screen power, network power and base power. Time-to-empty (TTE) is defined using the practical voltage collapse rather than the SOC to zero assumption. The model is assessed via local and global sensitivity analysis, and power-saving strategies are derived using an AHP multi-criteria decision-making framework. The SOC fitting quality reaches , and rank-correlation-based importance analysis identifies CPU-related workload factors as the dominant contributor to endurance variation, with a normalized importance score of approximately 40%. The model is evaluated using a train/test-separated validation protocol rather than relying only on fitting quality. Prediction errors are reported separately for SOC, terminal voltage, temperature, and voltage-cutoff-defined TTE. On the unseen test segments, the proposed model achieves SOC RMSE of 0.0402, terminal-voltage RMSE of 0.162 V, temperature RMSE of 1.954 K, and TTE RMSE of 0.34 h under the controlled simulation-based validation setting. These findings support strategies that prioritize CPU-load reduction and usage-aware control, and motivate voltage-collapse-aware power management for heavy workloads and aged batteries. Overall, the main message of this work is that reliable smartphone TTE prediction requires voltage-collapse-aware modeling rather than SOC-only extrapolation.
Yang et al. (Fri,) studied this question.