Multi-agent reinforcement learning for adaptive demand response in smart cities

Buildings account for over 70% of the electricity use in the US. As cities grow, high peaks of electricity consumption are becoming more frequent, which leads to higher prices for electricity. Demand response is the coordination of electrical loads such that they react to price signals and coordinate with each other to shave the peaks of electricity consumption. We explore the use of multi-agent deep deterministic policy gradient (DDPG), an adaptive and model-free reinforcement learning control algorithm, for coordination of several buildings in a demand response scenario. We conduct our experiment in a simulated environment with 10 buildings.