A Sensor-Aware Physics-Based Framework for Continuous Smartphone Battery State and Lifetime Prediction

Area of Science:

• * Mobile computing and energy management.
• * Battery modeling and state estimation.

Background:

• * Accurate prediction of battery state-of-charge (SOC) and time-to-empty (TTE) is crucial for mobile devices.
• * Existing models often lack the granularity to capture complex battery depletion behaviors influenced by multiple factors.

Purpose of the Study:

• * To develop a continuous-time, physics-based framework for predicting smartphone battery SOC and TTE.
• * To incorporate multi-source sensing and device operating information for enhanced prediction accuracy.
• * To provide an interpretable analysis of factors affecting battery consumption.

Main Methods:

• * Utilized a first-order RC equivalent circuit model, extended with six key factors: CPU workload, GPS, screen brightness, network, temperature, and battery aging.
• * Developed a multi-factor coupled SOC model using the GreenHub dataset.
• * Employed K-means clustering and the fourth-order Runge-Kutta algorithm for TTE prediction.
• * Introduced an interpretable factor analysis framework (sensitivity coefficients, partial contribution rates, entropy weights) to quantify factor impacts.

Main Results:

• * Achieved Mean Absolute Error (MAE) of 4.2% and Root Mean Square Error (RMSE) of 6.8% for SOC prediction.
• * Identified CPU workload as the dominant factor in battery depletion.
• * Confirmed significant impacts from GPS activity and screen usage on battery consumption.

Conclusions:

• * The proposed framework offers an interpretable and accurate solution for battery state perception in smartphones.
• * Findings provide valuable insights for optimizing energy management strategies in mobile sensing systems.
• * The physics-based approach enhances the understanding of battery dynamics under real-world usage conditions.