A study published in Discover Applied Sciences presents a Q-learning-based Dual-Battery Energy Management System (DBEMS) designed to optimize power allocation for agricultural electric vehicles. The research addresses the challenge of managing two lithium-ion battery packs under variable field conditions, where load demands can fluctuate significantly.
The study was conducted by Palla Siva Kumari and Mareddy Padma Lalitha, who developed a control strategy to handle the electrical behavior of the battery packs. The electrical behaviour of each battery is represented using a first-order resistor-capacitor model. The researchers subjected the system to various load profiles, including ploughing, transportation, and mixed-duty scenarios, to test the robustness of the energy management strategy under realistic agricultural conditions.
To evaluate the performance of the proposed controller, the team introduced Gaussian disturbances at levels of 0%, 5%, and 10% to simulate load uncertainty. The evaluation framework comprised 243 simulation runs, covering three different controllers, three load profiles, three noise levels, and nine random initializations.
All tested controllers successfully supplied the required load energy of 730 Wh, indicating that basic functionality was maintained across all strategies. However, the proposed Q-learning-based DBEMS achieved the highest Balance Index of 0.991. This result represents a 2.69% improvement over the rule-based controller, demonstrating a superior ability to keep the batteries balanced during operation.
The proposed controller reduced the switching frequency from 15 to 10 switches per kilowatt-hour, corresponding to a 33.33% reduction.
When compared to a conventional reinforcement learning benchmark, the proposed controller showed significant improvements in state-of-charge management. The mean State-of-Charge gap was reduced from 1.75% to 1.45%, a decrease of 17.14%. Additionally, the root-mean-square error dropped from 1.65% to 1.38%, representing a 16.36% reduction. These metrics indicate that the adaptive Q-learning method provides more precise control over the energy distribution between the two battery packs.
Under conditions of 10% load uncertainty, the controller retained stable performance, with a Balance Index standard deviation of approximately ±0.005. The variation in switching frequency remained below ±1.2 switches per kilowatt-hour, further confirming the robustness of the system.
The study highlights the effectiveness of adaptive reinforcement learning in managing dual-battery systems for electric vehicles used in agriculture.
Subject of Research: Adaptive reinforcement learning for dual-battery energy management in agricultural electric vehicles
Article Title: Adaptive reinforcement learning based dual battery energy management for agricultural electric vehicles under stochastic field load uncertainty
Article References: Kumari, P. S., & Lalitha, M. P. (2026). Adaptive reinforcement learning based dual battery energy management for agricultural electric vehicles under stochastic field load uncertainty. Discover Applied Sciences. https://doi.org/10.1007/s42452-026-09755-y
Image Credits: AI Generated
DOI: 10.1007/s42452-026-09755-y
Keywords: Q-learning, Dual-Battery Energy Management System, Agricultural Electric Vehicles, Lithium-ion Batteries, Stochastic Load Uncertainty, Reinforcement Learning, Battery Equalization, Switching Frequency, State-of-Charge, Resistor-Capacitor Model, Gaussian Disturbances, Simulation Runs, Balance Index, Root-Mean-Square Error, Energy Allocation, Agricultural Duty Cycles, Power Allocation, Control Strategy, Operational Efficiency, Battery Longevity
News Source: Faith Mcneil. (November 11, 2026). Q-learning improves battery balance in farm electric vehicles. Scienmag.



