As an emerging load category of an urban complex, load forecasting accuracy directly affects the planning and safe operation of the power grid. However, the load pattern of the urban complex is prone to change abnormally because of the influence of the external environment, and the accuracy of direct prediction cannot meet the requirements of actual operation. Therefore, it is necessary to cluster the load of the urban complex to extract different load patterns to improve the accuracy of prediction. This paper proposes a load pattern clustering method based on density-based spatial clustering and K-shape. First, the DBSCAN algorithm is used to extract the typical daily load curve of the complex load in different seasons according to the density of other regions. Then a K-shape clustering algorithm is used to cluster the typical daily load curves of different complexes in different seasons. Finally, the simulation results are compared with the clustering results of K-means and K-medoids. The results show that, compared with the other two methods, the two-stage load pattern clustering method of DBSCAN-K-shape proposed in this paper has high precision under different clustering indexes for urban complex load clustering. This work is supported by the Science and Technology Project of State Grid Corporation of China (No. SGGSJY00PSJS1900123).