The invention provides an electricity customer short-term load demand forecasting method. The method comprises the following steps: carrying out clustering on collected historical daily load curve data of electricity customers according to dates; then, establishing a load forecasting model at each time for each date group obtained through clustering; and finally, searching a historical similar day matched with a date to be forecasted, and obtaining a load forecasting result of the date to be forecasted according to the load forecasting model of the date group where the historical similar day belongs. According to the scheme, a parallel computation framework is utilized, and electricity load demands of mass electricity customers can be forecasted simultaneously; and requirements of mass data analysis speed and prediction accuracy are met.