This paper presents a nonintrusive appliance load monitoring (NILM) strategy for energy management systems in smart buildings. Compared to purely data-driven methods, this paper introduces a prior-knowledge-based model-driven framework. The prior-knowledge includes front-end power supply circuit identification, electrical operating principle, and customer usage. The focus of this paper is on a comprehensive study on front-end power supply circuit topologies and on the commercial power supply market. The former forms the basis of the proposed hierarchical taxonomy from the power electronics' point of view, and the latter study guarantees that the proposed taxonomy represents the majority of appliance loads in the real world. Under the proposed taxonomy, the advantage of the model-driven hierarchical feature extraction is discussed, and initial analysis shows that optimized features can be obtained from this method, which drives a much simpler AND more feasible solution in differentiating the subtle differences between similar loads.