A Case Study on Blending to Improve Day-ahead Forecasts of Residual Demand in Kyushu, Japan

Abstract

Residual power demand is the demand left to be met by conventional power generators after considering the power generation of renewable energy systems. Knowing this quantity ahead of time can be useful in the planning and operation of power systems in markets with high penetration of variable renewable energy. In this study we investigate the potential to improve day-ahead forecasts of residual power demand one day ahead of time through three blending techniques tested over two years of data for the balancing area of Kyushu in Japan. The results show that averaging can improve the root mean square error normalized by peak value by 3% over the best individual forecasting method. The maximum improvement achievable with blending was also identified as reaching 20% over the same reference. Finally, an expert-based method yielded results slightly better than averaging, but further pathways for its improvement were discussed and identified.