Predicting peak day and peak hour of electricity demand with ensemble machine learning

Battery energy storage systems can be used for peak demand reduction in power systems, leading to significant economic benefits. Two practical challenges are 1) accurately determining the load days and hours 2) quantifying reducing uncertainties associated with forecast probabilistic risk measures dispatch decision-making. In this study, we develop a supervised machine learning approach generate probability of next operation day containing hour month an day. Guidance is provided on preparation augmentation data as well selection models decision-making thresholds. The proposed applied Duke Energy Progress system successfully captures 69 out 72 testing months 3% exceedance threshold. On 90% days, actual among 2 h highest probabilities.