Since photovoltaic system (PVS) power outputs are not stationary, the persistence model, which assumes the last measured value will persist in the future, is normally applied to intra-hour forecasting. The persistence model with the extraterrestrial radiation (method A) is one of solving methods and can be applied at longer time horizons. However, the forecasting errors of it depend on installation conditions of PVS. To solve this problem, we developed the persistence model with maximum power outputs (method B). We compared the two methods in terms of the forecasting errors using irradiation data (METPV-11) and measuring data of 6 types of PVS power outputs. We found that on average during the evaluation period, method B is better than method A, and on clear sky days the advantage of method B becomes remarkable, however on cloudy days it declines and sometimes method B is worse than method A.