State of charge estimation for a UAV battery via Adaptive Extended Kalman Filter and Deep Reinforcement Learning
Year: 2026
Type: article
Abstract: This paper concerns with the development of an accurate battery State of Charge (SoC) estimation algorithm for mission/safety-critical applications. This problem is already challenging due to the nonlinear and flat regions in the Open Circuit Voltage-State of Charge (OCV-SoC) curve observed in many lithium-based batteries. However main difficulty in aviation comes from the environmental conditions battery being exposed; altitude that can reach to 55000 feet, temperature that can vary from \\-55$^{\circ}$ to +80$^{\circ}$, etc. This paper scrutinizes Adaptive Extended Kalman Filter (AEKF) based on 2RC-3RC models that of 1RC model that falls in short reflecting battery's fully-fledged characteristics, while other varying conditions are regarded with Deep Reinforcement Learning (DRL) approach. The DRL approach is characterized by its dynamic adjustment of filter parameters, which is achieved through the learning of optimal covariance values in real time. In the proposed method, Process Covariance Matrix (\(Q_p\)) and the Measurement Covariance Matrix (\(R_m\)) are statistically estimated using Proximal Policy Optimization (PPO)-based policy learning within a clipped adaptive range to leverage the stability and flexibility of policy learning in PPO to accurate SoC estimation for safe flights. The proposed method's performance evaluation is verified on a UAV battery in a real flight duty.
Institutions: Istanbul Technical University, Ankara Medipol Üniversitesi
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,800
Funder: Türk Havacılık ve Uzay Sanayii
OpenAlex ID: W7147254208