Residential Electricity Rate Plans and Their Selections Based on Statistical Learning

Demand response (DR) is one of the major benefits that utility companies can derive from the advanced metering infrastructure (AMI). In particular, the dynamic rate plan with DR is attracting attention as an electricity rate system suitable for the future power environment. In order for electricity consumers to select an appropriate electricity rate plan, it is necessary to provide information such as whether electricity bills are reduced by the plan and the estimated amount of electricity bill savings. In this paper, we first comparatively analyze the current progressive rate plan and a dynamic rate plan of the time-of-use (TOU). We next propose several prediction methods for households to provide information on whether the electricity bill amount can be reduced in advance when changing to the TOU rate plan from the progressive rate plan by using only the current monthly electricity usages and bills. In order to develop three different prediction methods based on statistical learning, we use the support vector machine, linear regression, and deep neural network techniques. As a ground truth training sequence, we use hourly electricity usages and bills obtained from ten apartment complexes through AMI, and an apartment complex is used for testing the designed methods. The decision accuracy for the test complex was more than 0.98 and the root mean square error of the saving prediction was 1.7%.