With the increasing proportion of new energy power generation, the adjustable capacity of the generation side is reduced. Mining the adjustable potential of the load side is an inevitable choice for the balance of power supply and demand in the new power system. A feature extraction and comprehensive evaluation method of adjustable potential for industrial users based on secondary clustering is proposed. Taking into account the interruptibility, transferability and production attributes, an evaluation index system of load adjustable potential is built. First, the secondary clustering method of affinity propagation clustering, k-means clustering, is proposed to mine adjustable potential features and calculate adjustable indices. Second, the methods of criteria importance through intercriteria correlation (CRITIC) and vlsekriterijumska optimizacija I kompromisno resenje (VIKOR) are used to evaluate the interruptible, transferable and production potentials of industrial users. Then, the set pair analysis-variable fuzzy set (SPA-VFS) method is used to evaluate the comprehensive adjustable potential of industrial users and to obtain the priority ranking. Finally, the adjustable potential of nine industrial users is evaluated using historical load monitoring data. By comparing different clustering methods, index systems and evaluation algorithms, the advantages of the proposed method in feature extraction, index definition and evaluation rationality are verified.