The emerging of energy internet calls for even higher requirements for the plant-level load dispatching of thermal power units in terms of safety, efficiency, cleanness and intelligence. The physical system, information system and control system of power units were supposed to deeply fused, coordinated and optimized. The cyber-physical system (CPS) integrates the physical, communication and information network, providing a new way for the intelligent load dispatching of thermal power units. With the large amount of operational data, FRS-based big data analytics was introduced to derive the correlation between the physical and information model of coal consumption and pollutant emission; considering the economic and pollutant emission factors, the CP-based load dispatching model was built. The model was simulated on the operation data of two 600MW coal-fired power units, and the results show that the reduction range of coal consumption rate is 1.3g/(kW?h), and the concentration reduction range of SO2, NOx and dust is 3.4mg/m3, 2.5mg/m3 and 1.6mg/m3, respectively.