An Optimal Charging and Discharging Scheduling Algorithm of Energy Storage System to Save Electricity Pricing Using Reinforcement Learning in Urban Railway System

초록

This paper proposes the optimal charging and discharging scheduling algorithm of energy storage systems based on reinforcement learning to save electricity pricing of an urban railway system in Korea. Optimization is done through reinforcement learning of charging and discharging schedule of energy storage systems according to the unit of electricity pricing rates as well as a reduction of peak power demand to save electricity pricing. To do this, modeling of urban railway systems including energy storage systems, electricity pricing rates, and changes in rates according to operations of energy storage systems are carried out. Reinforcement learning for an agent is also done to reduce peak power demand through DQN algorithm. Operation data of actual lines of urban railways operating with energy storage systems are utilized for learning. For this reinforcement learning, about 399(45.3%) incorrect data are removed and 481(54.7%) normal data are extracted. Through the reinforcement learning, maximum peak power demand is reduced by a targeted amount, 100 kW, from 2,982.4 kW to 2,882.4 kW. When the peak power demand is under 2,600 kW, charging at times when the power rate is cheaper and discharging at times when the power rate is more expensive are carried out, thus saving the total electricity pricing.

키워드