Comprehensive rural distribution network optimization: Tailored demand-side management via multi-agent deep reinforcement learning coupled with distributionally robust stochastic models

The increasing penetration of renewable energy in rural distribution networks presents a critical opportunity to transition toward sustainable energy systems. However, rural networks face unique challenges, including geographical dispersion, intermittent renewable generation, and socio-economic constraints, which complicate effective energy management. To address these issues, this paper proposes a novel Adaptive Demand-Side Management (DSM) framework tailored for rural distribution networks with high renewable energy integration. The framework integrates Multi-Agent Deep Reinforcement Learning (MADRL) with Distributionally Robust Optimization (DRO) to enable decentralized, adaptive, and resilient decision-making under uncertain conditions. The MADRL component models distributed energy resources (DERs), such as renewable generators, storage systems, and flexible loads. The proposed framework is validated through extensive simulations on a rural distribution network case study. Results demonstrate significant improvements in renewable energy utilization, voltage stability, and overall system resilience. Key findings include a 20% increase in renewable energy utilization, a 15% reduction in voltage deviations, and enhanced adaptability to variable load and generation conditions. This research contributes to the growing body of knowledge on DSM in rural energy systems, offering a scalable and robust solution to support the global transition toward low-carbon, sustainable energy infrastructures.