Reinforcement Learning for Energy Storage Optimization in the Smart Grid

In the existing electrical grid, there is a large demand supply mismatch which results in unutilized power potential. A demand based pricing approach can be achieved by implementing dynamic pricing in the smart grid. But such an approach can only be effective when the consumer has a mechanism to adapt their power usage based on the price of power. This paper looks into the implementation of Reinforcement Learning algorithms- specifically, Q-learning and SARSA [1] - to control batteries to optimize energy storage at a larger scale. We also demonstrate a non-linear algorithm for “bucketizing” which allows the incorporation of states with highly uneven distributions of visit frequency into the model.