Battery State of Charge Estimation with High Accuracy Coulomb Counting

<p dir="ltr">State of Charge estimation is a critical function in battery management systems directly influencing power management and safety in applications such as drones. While the Coulomb Counting Method is widely adopted for its low computational complexity, it is inherently susceptible to cumulative drift caused by sensor noise and numerical integration errors. This thesis investigates the impact of replacing traditional first-order Forward Euler integration with a Fourth-Order Runge-Kutta method to improve State of Charge estimation accuracy. The study evaluates the numerical performance of both schemes using synthetic current profiles designed to emulate various signal dynamics, followed by validation with real-world experimental UAV flight data. Synthetic results demonstrate that while the Euler method exhibits systemic, frequency-dependent divergence, the Runge Kutta fourth order method reduces discretization errors, achieving near-zero absolute SOC error under controlled conditions. Experimental validation using a quadrotor platform in stabilized flight shows that the Runge Kutta method consistently maintains a lower error profile than the Euler method, achieving an approximately 13.75% reduction in mean absolute error for one charge/discharge cycle. In practice, there might be many cycles before recalibration is possible, thus, a small reduction in estimation error may make a big difference in the long run, making this a key improvement in battery charge management. These findings suggest that higher order numerical integration serves as a low-complexity software-level optimization that enhances the reliability of SOC estimation without requiring additional sensing hardware or complex battery models. Despite current experimental limitations related to onboard sensor accuracy and project time constraints, this research provides a potential path for improving power management systems in high-dynamic applications.</p>