Verma et al. Res. Trends Int. J. Technol. Innov., January - March 2026, 1 (1) : 36-44
1International Institute of Information Technology Bangalore, Bangalore, India; 2Department of Computer Applications, Amrita Vishwa Vidyapeetham, Coimbatore, India
Article History
Accepted : 02 Mar 2026
Published : 28 Mar 2026
Publication Issue
Volume 1, Issue 1
January - March 2026
Page Number36–44
Hybrid renewable microgrids combining solar, wind and battery storage give operators flexibility but complicate dispatch decisions that must jointly consider unmet-demand risk and heterogeneous generation costs. This paper formulates the dispatch problem as a Markov decision process and trains a deep Q-network agent to decide generation-source and battery dispatch across a 12-node microgrid benchmark. Compared with threshold-based rule dispatch, the learned policy reduced levelised energy cost by 27 percent while keeping unmet-demand events under the 2 percent target.
Keywords - reinforcement learning, hybrid microgrid, renewable energy dispatch, cost optimization, energy storage
Rule-based microgrid dispatch controllers react to state-of-charge or generation thresholds but ignore the cost differential between distributed generation sources and battery cycling wear, often over-relying on the grid tie-in during peak-price windows. This work explores whether a learned dispatch policy can jointly optimise for cost and reliability across generation sources.
A 12-node hybrid microgrid benchmark (based on an open-source distribution feeder reference model with solar PV, wind and battery storage nodes) was simulated across a full annual load profile. A deep Q-network was trained over 3,000 simulated episodes using historical generation and demand traces, with state features including per-node demand, current battery state of charge, and spot grid-tariff pricing, and actions corresponding to dispatch from solar, wind, battery or grid-tie sources.
The DQN policy reduced average levelised energy cost by 27 percent relative to threshold-based rule dispatch over a two-week evaluation window replaying representative demand traces, while unmet-demand events remained at 1.6 percent, below the 2 percent target. Convergence was reached after approximately 1,800 training episodes.
Reinforcement learning provides a viable mechanism for cost-aware dispatch decisions in hybrid renewable microgrids without manual threshold tuning. Future work will incorporate carbon-intensity signals as an additional optimisation objective alongside cost and reliability.
[1] Mnih V. et al., Human-level control through deep reinforcement learning, Nature, 2015. [2] Olivares D. E. et al., Trends in microgrid control, IEEE Transactions on Smart Grid, 2014. [3] Zia M. F. et al., Microgrids energy management systems: A critical review on methods, solutions and prospects, Applied Energy, 2018.
© 2026 The Author(s). Published by IJEIA Editorial Office. This is an open access article under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
Aditya Verma, Lakshmi Narayanan (2026). Cost-Aware Dispatch Scheduling for Hybrid Renewable Microgrids Using Reinforcement Learning. IJEIA, 1(1), 36-44.