Separating Instantaneous and Delayed Battery-Drain Effects for Smartphone Remaining Runtime Prediction

Smartphone time-to-empty (TTE) estimation is vulnerable to residual battery-drain effects after short-duration user activities, which can cause runtime predictions to fluctuate. This paper proposes a continuous-time battery-drain model that separates instantaneous and delayed drain effects in operating-system logs. The discharge process is decomposed into baseline drain, direct drain induced by the current usage state, and delayed tail drain triggered by transient events. Multi-exponential recursive states are used to represent network communication, location requests, background wake-ups, and related events as residual drain terms that decay over time. Non-negative sparse estimation is adopted to preserve the interpretability of the individual contributions. Experiments are conducted on a public smartphone-log dataset with strict separation between the training and test sets. The results show that the proposed method achieves stable TTE prediction performance on the test set, with short-horizon R2 values above 0.963 and scenario-level MAPE of 2.42–5.03%. These findings indicate that explicit modeling of delayed residual drain can improve the accuracy and stability of smartphone remaining-runtime prediction.