Recently, with the growing PV capacity, analyzing its impacts on reliability of power grid accurately has practical significance.This paper puts forward a novel method for modeling the PV output, based on optimal segmentation and multidimensional clustering method, which can retention circadian and seasonal timing characteristics well.Firstly, according to the effective output time of PV in a day, using Fisher optimal segmentation algorithm, year simulation period gets optimal segmentation.Secondly, using multidimensional cluster method of fuzzy c-mean cluster algorithm, each period is analyzed.This method makes photovoltaic output modeling more refined, and preserves its local properties better.Applied in IEEE-RTS test system, the reliability analysis uses meas-ured PV output in Ningxia.The year simulation period is divided into five optimal periods, and at each period, the impact on power system reliability of PV is analyzed.The result shows the effectiveness of the method, and provides significant guidance on PV capacity planning.