Minimizing Reactive Power Sharing Error in Islanded Microgrids via SADRL-TD3 Adaptive Hybrid Droop Control
Year: 2026
Type: conference-paper
Abstract: Reactive power sharing in islanded inverter-based microgrids is often inaccurate due to mismatches in line impedance and other non-idealities. Such poor reactive power sharing can lead to circulating currents and system instability. This paper proposes a single-agent deep reinforcement learning (SADRL) approach to adapt the Q–V droop coefficient in real time for a two-inverter islanded microgrid. The agent is trained with the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm and utilizes local measurements of voltage, frequency, and real/reactive power errors as state inputs. By continuously tuning the reactive power–voltage droop coefficient, the SADRL controller minimizes the reactive power sharing error while maintaining stable voltage and frequency and ensuring acceptable active power distribution. The proposed method is implemented in a high-fidelity MATLAB/Simulink simulation for validation. Results demonstrate that the SADRL-TD3 adaptive hybrid droop controller significantly reduces reactive power sharing errors compared to a conventional fixed-droop control. The microgrid achieves a more balanced reactive power distribution between the two inverters, with voltage and frequency deviations kept within acceptable limits. These results highlight the potential of deep reinforcement learning for enhancing primary control in microgrids, achieving superior power sharing performance under steady and dynamic conditions.
Institution: JSPM University
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204937795