Title: Grid load balancing with V2G and deep reinforcement learning

Authors: R. Sasirega; S. Prakash

Addresses: Department of EEE, Bharath Institute of Higher Education and Research (BIHER), Chennai, India ' Department of EEE, Bharath Institute of Higher Education and Research (BIHER), Chennai, India

Abstract: Charging and discharging maximisation of electric vehicles remains a challenge primarily attributed to grid condition changes and different energy requirements. This paper presents an innovative solution to this problem using deep reinforcement learning, which changes the charging and discharging schedules of electric vehicles depending on the real time frame. A hybrid feature selection method that merges the Pelican Optimisation Algorithm (POA) and Gannet optimisation algorithm (GOA) is utilised to select the relevant features necessary for effective training of the proposed model. For optimal charging and discharging rates, the theoretical concepts of the Deep Q-Learning (DQL) algorithm were incorporated. The proposed approach delivered impressive results, with a peak demand of just 110 kW and a total energy usage of 3200 kWh, showcasing its stellar performance in managing the grid load. Moreover, it boasted a grid stability index of 0.92, converged in 20 seconds, and achieved an outstanding model accuracy of 95%.

Keywords: deep Q-learning; EV charging/discharging; V2G technology; optimisation; feature selection.

DOI: 10.1504/IJWMC.2026.155712

International Journal of Wireless and Mobile Computing, 2026 Vol.31 No.2, pp.141 - 153

Received: 28 Jun 2024
Accepted: 09 Jan 2025
Published online: 11 Aug 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article