The Effect of Temporal Aggregation on Battery Sizing for Peak Shaving

Battery systems can reduce the peak electrical consumption through proper charging and discharging strategies. To this end, consumers often rely on historic consumption data to select a cost-efficient battery system. However, historic data is an imperfect mapping of the real consumption, because of a coarse sampling rate or measurement inaccuracies. This can result in non-optimal decisions, e.g., by underestimating the battery capacity required. In this article, we analyze how aggregation affects a state-of-the-art battery sizing algorithm for an industrial production site. We then use machine learning on a short period of high-resolution data to correct this error from historic data. Our experiments indicate that machine learning models can correct this error in some cases. However, adding a safety margin obtained from historic data to the battery size is a more reliable way of reducing the error.