Coordinated Scheduling of Flexible Resources in Active Distribution Networks Using Multi-Agent Reinforcement Learning and Safety-Constrained Control

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

The rapid popularization of controllable loads such as electric vehicles (EVs), battery energy storage systems (BESS), and heating, ventilating and air conditioning (HVAC) has greatly improved the flexibility of active distribution networks, but it also brings challenges to safe and efficient real-time scheduling. However, most existing control strategies rely on either static optimization or single-agent reinforcement learning, which have shortcomings in scalability, adaptability to dynamic operating conditions, and ensuring network constraint satisfaction. To remedy these deficiencies, we propose a multi-agent reinforcement learning (MARL) -based scheduling framework combined with quadratic programming (QP) safety projection, which unifies flexible resource modeling, learns adaptive cost-aware and comfort-aware control policies, and guarantees real-time physical feasibility. The case study on the IEEE 123-node test system shows the superior performance of the proposed method compared to the existing methods. Specifically, in a typical scenario with 40% PV penetration, the proposed framework achieves 19.3% peak demand reduction and 13.4% operational cost savings. Moreover, it maintains a high EV user satisfaction of 96.5% and strictly guarantees network security with a zero voltage violation rate, which indicates its potential for reliable deployment in modern active distribution networks.

Links and sources

• PDF from original source

Comments