Abstract: Non-intrusive load monitoring is an effective way to obtain load data and realize load perception. A combined support vector machine identification method for making the process of the non-intrusive load monitoring universal and practical based on the structured characteristic spectrum, is studied to automatically execute the process without disturbing users and achieve high identification accuracy. By constructing the characteristic spectrum of typical loads, the disordered waveform data are transformed into structured signature data, which makes the signature graph universal and distinguishable. Based on the structured characteristic spectrum, the support vector machine classifier model of typical loads is established, and the combined support vector machine classifier of each type of load is formed based on the base classifier model. The idea of“gathering weak into strong”is used to ensure that each combined classifier has high classification accuracy, so as to realize the accurate load identification. Based on the universal graph and the classifier model, the real-time non-intrusive load identification can be realized through the processing flow of event waveform extraction, waveform data structuration and classifier decision. By the acquired actual load data for verification,the characteristic spectrum of the typical load is developed, the collected data of multiple households based on the combined support vector machine model are classified, and the high accuracy identification for the load data of different users is achieved,which verifies that the proposed method has good universality and effectiveness.