Abstract
Reinforcement learning is a promising model-free and adaptive controller fordemand side management, as part of the future smart grid, at the districtlevel. This paper presents the results of the algorithm that was submitted forthe CityLearn Challenge, which was hosted in early 2020 with the aim ofdesigning and tuning a reinforcement learning agent to flatten and smooth theaggregated curve of electrical demand of a district of diverse buildings. Theproposed solution secured second place in the challenge using a centralised'Soft Actor Critic' deep reinforcement learning agent that was able to handlecontinuous action spaces. The controller was able to achieve an averaged scoreof 0.967 on the challenge dataset comprising of different buildings andclimates. This highlights the potential application of deep reinforcementlearning as a plug-and-play style controller, that is capable of handlingdifferent climates and a heterogenous building stock, for district demand sidemanagement of buildings.