Provided are a power grid short-term load forecasting method and system. A daily load curve variation coefficient of a forecasted day is calculated according to historical data of dates in the type similar to that of the forecasted day; regression analysis is conducted according to the historical data of the dates in the same type to obtain a regression equation of the maximum daily load and the highest daily temperature and a regression equation of the minimum daily load and the lowest daily temperature; the maximum load and the minimum load of the forecasted day are forecasted according to the relation between the maximum daily load and the highest daily temperature and the relation between the minimum daily load and the lowest daily temperature, and forecasted load values of various durations of the forecasted day are obtained in combination with the daily load curve variation coefficient of the forecasted day. Because influences of temperature changes on daily load curve variation are considered, short-term load forecasting precision is improved under special air temperature situations such as hot summer, and compared with a traditional power grid short-term load forecasting method, the method improves forecasting accuracy.