A Deep Reinforcement Learning- and Linear Programming-based Hierarchical Aggregation Framework for Demand-side Flexibility

This paper proposes a scalable HVAC hierarchical aggregation method that integrates Deep Reinforcement Learning (DRL) and Linear Programming (LP) to adaptively balance demand-side flexibility and thermal comfort.In the proposal, DRL determines the flexibility requirement commands issued to aggregators, while LP optimizes HVAC control commands to generate flexibility while maintaining thermal comfort.Our approach ensures scalability and adaptability in large-scale demand-side management.The simulation results demonstrate that the proposed method successfully manages HVAC operations by up to 600 rooms, achieving the required flexibility without violating comfort.