Withing the development of power industry and the continuous improvement of electric power metering system，the features of power grid demand side present a diversified development． Smart meters come into people＇s life，which is bringing huge amounts of electricity data． A meaningful problem is how to explore the behavior of users to promote the electricity market． Combining the considerations of traditional K － modes method and hierarchical K － means method，moving hierarchical K － modes method is proposed，which is able to deal with categorical data and gives a reasonable k value of the number of clusters． Moreover，a method to transform curve data into categorical data by differencing and categorizing is proposed，which can preferably reflect the shape of curves． At last，clusetering experiment is carried out by using moving hierarchical K － modes method based on simulated data and real data of power users in Xiamen island region，which gets excellent clustering results．