Power disaggregation that breaks down the overall power consumption to appliance level acts as a feasible technical solution to meet the extensive data demand but reduce the costs of installing advanced metering system in Demand Side Management (DSM). Considering the intensive query of high-frequency training data of existing methods, this paper presents a new behavior based model applicable to low-frequency data by introducing external determining factors of power consumption into a finite mixture model (FMM) that disaggregates overall power consumption into those of various electrical appliances. Empirical verification by employing a dataset including detailed hourly appliance-level power consumption of commercial buildings in Shanghai proves that this newly developed model can provide more accurate result than other previous models but requires relatively lower-frequency data. The benefits of energy-saving potential from information feedback and appliance replacement facilitated by disaggregation data is further simulated to show the practical application.