Sequential service restoration with grid-interactive flexibility from building AC systems for resilient microgrids under endogenous and exogenous uncertainties

Microgrids can operate in island mode and can restore critical loads utilizing renewable energy sources (RESs) after natural disasters. Yet, the uncertainty of RESs poses a significant challenge to their participation in the microgrid sequential service restoration (SR) process. Air conditioning (AC) systems with building thermal inertia can provide grid-interactive flexibility by pre-cooling or pre-heating strategies while satisfying occupants’ comfort level, and are ideal resources to handle uncertain renewable generation. However, endogenous uncertainty which captures uncertain time-varying grid-interactive flexibility realization of building AC systems affected by load pickup decisions, together with exogenous uncertainty from renewable generation, load demand, and outdoor ambient temperature make the problem hard to solve. To address above-mentioned challenges, a deep reinforcement learning-based (DRL) resilient microgrid SR approach coordinating high penetration of RESs and grid-interactive flexibility from building AC systems is proposed in this work. The microgrid SR problem is reformulated as a Markov decision process (MDP) and addressed with an improved twin delayed deep deterministic policy gradient-based (TD3) microgrid SR algorithm. The proposed microgrid SR approach is verified on the IEEE 33-node and 123-node systems. Comprehensive numerical results demonstrate that the proposed microgrid SR strategy with grid-interactive flexibility from building AC systems can attain superior performance in terms of restoration performance indices.