The invention provides a rapid power user classification method aimed at the peak-valley difference of a load curve. The method comprises the following steps that 1) daily load data of power users is obtained; 2) the data is normalized; 3) a corresponding peak-valley period is obtained; and 4) according to a correlation coefficient function, iteration is repeated, and a power user classification result similar to the peak-valley period curve is finally obtained. The power users can be clustered according to the daily load curve data of the users, and power companies can make strategies in a targeted way subsequently to reduce the peak-valley difference.