The opening electricity market and the incremental penetration of renewable energy provide more consumption choices for users. This results in diversification of power user patterns, increasing differences of load characteristics and giving a complex distribution of load clusters. An improved self-adaptive density peak clustering (ISDPC) algorithm is proposed to ameliorate the clustering results and adaptive abilities of traditional clustering methods for unbalanced load data. First, a new density metric is defined based on the K-nearest neighbor (KNN) and relative density. Secondly, the optimal number of clusters is obtained by a fitting partition function obtained from the decision graph. Finally, the allocation of strategy is improved by a weighted KNN graph. The experimental results show that clustering results obtained from the proposed method perform better in accuracy, robustness, and adaptability. This work is supported by the National Natural Science Foundation of China (No. 51877181).