Economically Efficient Power Storage Operation by Dealing with the Non-Normality of Power Prediction

Abstract

Various predictive models about the residential energy demand and residential renewable energy production have been proposed. Recent studies have confirmed that they are not normally distributed over time. The increase in renewable energy installation has brought the issue of energy storage charge and discharge control. Thus, storage control methods that properly address non-normality are required. In this paper, we formulated the economically optimal storage control problem using Markov decision process (MDP) and the conditional value at risk (CVaR) measure to deal with the non-normality of predictive distribution about the household’s net load. The CVaR measure was employed to treat with the chance constraint on the battery capacitor, in other words, overcharge risk and over-discharge risk. We conducted a simulation to compare the annual economic saving performances between two MDPs: one is the MDP with a Gaussian predictive distribution and the other is the MDP with a normalized frequency distribution (non-normal). We used the real time series of 35 residential energy consumption and PV generation data in Japan. The importance of addressing the non-normality of random variables was shown by our simulation.