Answers are grounded in this paper and cite their sources.
This paper proposes a multi-agent deep deterministic policy gradient with prioritized experience replay and cooperative gradient sharing (MADDPG-PER-Coop) framework for automatic generation control (AGC) in renewable-integrated interconnected power systems. The proposed approach integrates prioritized experience replay, adaptive exploration noise, cooperative gradient sharing, State-of-Energy observation augmentation, and a composite multi-objective reward function to improve frequency regulation and tie-line power control under varying operating conditions. The framework is investigated using a three-area thermal–hydro–wind–solar photovoltaic interconnected power system and further examined in an extended five-area configuration to assess its performance under increased system dimensionality. Under the nominal benchmark condition, the proposed controller achieves a settling time of 8.0 s, an overshoot of 2.1%, an integral squared error (ISE) of (8.6×10⁻³), and an integral of time-weighted absolute error (ITAE) of 0.21. Compared with the conventional proportional–integral–derivative (PID) controller, these results correspond to reductions of 68.4%, 85.2%, 80.9%, and 76.4% in settling time, overshoot, ISE, and ITAE, respectively. In a separate stochastic evaluation involving 30 independent simulation runs, the proposed method achieves a mean ITAE of 0.248 with a standard deviation of 0.015 and a mean settling time of 5.94 s. The ablation analysis examines the contribution of the individual framework components to the observed control performance. In addition, real-time implementation using a dSPACE DS1104 platform yields a mean execution latency of 1.86 ms and a maximum latency of 2.47 ms, with peak frequency deviations of approximately 0.010–0.012 Hz across the evaluated areas. The results indicate that the proposed framework provides effective frequency regulation under the investigated simulation conditions and demonstrates real-time execution feasibility on the tested platform. The findings are limited to the evaluated system configurations and operating scenarios, and further assessment is required to establish its applicability to larger and more diverse power systems.
No takes yet. Share an insight, caveat, or question.
Krishnaveni et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: