Managing Load Uncertainty Leveraging Real-time Scheduling of Energy Storage System

In this paper, we propose a real time control method for energy storage system (ESS) to minimize the total cost including the loss caused by load prediction error. To minimize total cost, we consider three types of costs: the energy cost from time-of-use (TOU) tariff, the monthly base cost determined from peak power and battery degradation cost. In order to consider the state-dependent battery degradation cost, we use dynamic programming for the optimal policy. Meanwhile, under the Korea commercial and industrial tariff, minimizing the peak power consumption is critical because it determines the monthly base cost and furthermore may affect for the next 12 months. However, peak shaving is vulnerable to the errors from day-ahead load prediction, and thus conservative strategies are desired. To alleviate the performance loss incurred from the load prediction error, we propose a concept of marginal power and learn the relationship between the marginal power and the load prediction errors based on the historical data. By exploiting the marginal power, the real-time ESS charging/discharging power becomes close to the optimal offline policy. Our extensive simulation for one year shows that the proposed method reduces the peak power cost by 45%.