Effect of Day-ahead Forecasts on Curtailment Planning of PV Power in Japan
Abstract
The objective of this study is to evaluate the effect that the accuracy of day-ahead forecasts of photovoltaic, PV, power generation has on forecast based curtailment of PV power. Two main regions of Japan were targeted, Kanto, which has a high potential for installations of roof-top PV systems; and Kyushu, which already has high levels of PV power penetration. To provide a qualitative measure of the importance of accurate PV forecasts with different levels of PV power generation, for each region, 2 penetration scenarios were assumed. One year of regional and day-ahead forecasts of PV power were done using numerical weather prediction data, support vector regression, and measured PV power generation data. The PV power generation data comes from to a set of 52 PV systems in Kyushu and to a set of 62 PV systems located in Kanto. The curtailment of PV power was planned one day ahead of time, using the PV power forecasts and a method based on residual loads. The results show that with current day-ahead forecasts it is possible to correctly predict the hours when curtailment will occur most of the time. For example, for Kyushu, curtailment was properly detected 83.1% of the time in a scenario of 8.64 GW of PV installed. For Kanto the values were between 73% and 80.5%. Regarding the effect of the forecast error, we found that it annualy represents near to 35% of the amount curtailed, indicating the importance of improvement of day-ahead forecasts in the curtaiment problem. Finally, the study shows that the relation between the months and hours of curtailment with the period of high forecast errors is crutial to the efficient use of the forecasts, and must be considered before using the forecasts to plan curtailment.
Keywords
Photovoltaic power generation, Day-ahead regional forecasts, Curtailment planning, Support vector regression.