The invention relates to a short-period forecasting method for a photovoltaic output, and particularly to a short-period forecasting method for the photovoltaic output in consideration of an assembly temperature. According to the method of the invention, the assembly temperature as one factor which directly affects the photovoltaic output is used as an important consideration factor for forecasting the photovoltaic output. According to the short-period forecasting method, the factors which affect the assembly temperature mainly comprise radiation strength, ambient temperature, wind and humidity. For accurately forecasting a photovoltaic output power, the temperature of the photovoltaic assembly is forecasted by means of a GDPSO-LSSVM algorithm. An entropy weight method is used for selecting similar days, and furthermore entropy weight method is used for analyzing different photovoltaic characteristic indexes and outputting a variation degree in the power influence. The photovoltaic output power is forecasted by means of a computed information entropy and the GDPSO-LSSVM algorithm.