An Adaptive Multi-Layer Heuristic Framework for Real-Time Energy Optimization in Smart Grids

Smart grids face significant challenges in coordinating demand-side management (DSM), dynamic pricing, data aggregation, and network feasibility in real time. To address this, we propose H-EMOS-Lite, an adaptive, multi-layer heuristic framework that integrates these components into a unified, real-time optimization loop. Evaluated on fully reproducible generated demand, price, and grid datasets based on realistic residential energy systems, H-EMOS-Lite achieves a 2.1% reduction in peak load and completes a full 24 h (96-interval) optimization for 100 households in under 0.25 s, demonstrating its suitability for near-real-time residential energy systems. The framework outperforms three baselines—Independent DSM, Sequential Optimization, and Particle Swarm Optimization (PSO)—by effectively balancing energy cost, peak load reduction, and temporal smoothness of the aggregate load profile, while avoiding abrupt, unsynchronized load shifts that induce secondary peaks—common in uncoordinated approaches. By embedding physical feasibility and cross-layer feedback directly into the optimization loop, H-EMOS-Lite enables scalable, interpretable, and deployable coordination for smart distribution systems.