CHARACTERISTICS OF DAY-AHEAD RESIDUAL DEMAND, PV POWER AND DEMAND FORECASTS IN A SCENARIO OF LARGE PENETRATION OF PV

摘要

The objective of this study is to characterize and compare day-ahead forecast errors of PV power generation and power demand in regional scale, and in a scenario of large penetration of PV power. Furthermore the effects of both forecast errors on a residual demand forecast are also investigated. The study targets the area with the highest penetration of PV power in Japan, Kyushu region. The forecasts of PV power and power demand were done with machine learning techniques and numerical weather prediction data. An analysis of forecasts of PV power indicates that its large forecast errors have an asymmetry of values according to the forecast value. Such asymmetry was not observed in the case of power demand. The results also show that PV power forecast errors seem to be uncorrelated with those of power demand forecasts. Regarding residual power demand forecasts, its errors had a variability resembling the one of power demand forecast errors and outliers approaching the values of those of PV power forecast errors. Forecasts for a 2-year period had RMSE of 0.059 GW/GW_(peak) for PV power forecasts, 0.028 GW/GW_(peak) for power demand forecasts and 0.041 GW/GW_(peak) for residual power demand forecasts.