Non-intrusive load identification for special transformer users is an important step to realize the demand-side response with special transformer users as the main subject. Through the feature analysis on the steady-state process and transient-state process of the special transformer users loads, the active power, reactive power, current effective value, etc., are selected as the characteristic quantities of the steady-state process. Moreover, the characteristic quantities before, during and after the occurrence of transient, such as current effective value, active power average, reactive power average, duration, maximum effective current value, etc., are selected to construct a comprehensive load feature space based on the steady-state process and the transient process for non-intrusive load identification. On this basis, the least square QR (LSQR) decomposition algorithm is used to decompose the steady-state composite load in order to obtain the operation of various loads. And the system distance clustering algorithm is proposed to classify transient events of the common industrial loads and further identify which type of loads has switched. Finally, the field data of industrial loads including drainage pumps, mixers, cement spirals, sewage pumps, dust collectors, etc., is collected and the simulation analysis of the proposed method is carried out. The results show that the non-intrusive load identification method proposed in this paper can achieve high accuracy.