The Japanese Society for Artificial Intelligence

Along with the spread of renewable energy on the customer side, the detailed management and forecasting of electricity demand for small groups has become a challenge for power transmission and distribution business operator. In general, demand forecasting predicts demand over a wide area based on weather forecasts. However, compared to demand forecasts for the area where the statistical properties of a relatively large number of consumers are bundled together and can be modelled, demand forecasts for smaller, more detailed groups are difficult to forecast because demand increases or decreases significantly depending on individual consumers. In this study, we focused on next-generation smart meters, which are scheduled to be introduced from 2025. We investigated whether demand forecasting for small groups can be made more accurate with the power data of main power of individual consumers and individual home appliances, which will be newly collectable by the next-generation smart meters. Experiments using electricity data from seven households showed that the accuracy of demand forecasting can be improved by using the electricity data of main power sources and appliances of individual consumers as features in addition to the past demand of the group.