Designing and validating lightweight machine-learning models that estimate battery State-of-Charge (SoC) in real time for low-power electronic devices operating under varying temperature and load conditions.
Overview
Knowing how much charge a battery really has left is deceptively hard: the answer shifts with temperature, load, ageing and chemistry. This project set out to estimate battery State-of-Charge (SoC) accurately and in real time for low-power electronic devices, using lightweight machine-learning models that are small enough to run on-device.
The work was a collaboration with Dr. Mohamad Khairi Ishak (Principal Investigator, Ajman University), building on a shared line of research in embedded intelligence and energy-aware systems.
What we did
- Built a battery monitoring and controlled-discharge rig to run repeatable experiments.
- Collected discharge data across a range of temperature and load profiles, capturing how real devices behave rather than idealised bench conditions.
- Compiled a clean, documented dataset and code repository that can be extended by future work.
- Designed and validated LSTM-based SoC estimators, tuned with Keras-Tuner and a Genetic-Algorithm-based optimisation, and tested them for on-device feasibility.
Outcomes
The project completed all of its objectives, producing an interpretable, deployable SoC estimation pipeline alongside an open, reusable dataset and codebase. Its results are reported in two peer-reviewed outputs — a conference paper (ICSECS 2025) and a journal article (Energy Reports, 2025) — both listed on the Research & Publications page. The accompanying code and data are openly available on OSF.
Assoc. Prof. Dr. Mohamad Khairi Ishak
Co-Investigator
University of Sharjah