Abstract: High-precision mapping model of the power plant system is the core technology of advanced scheduling and control in smart power plants. The traditional method of obtaining data through field tests is costly and time-consuming. Meanwhile, online mechanism model relies on expensive simulation software. This study proposed a construction method of data surrogate mapping model of the thermal power plant system based on mechanism simulation extension. Firstly, an offline structural mechanism model of the thermal system was established, and the model was identified and modified by the operating data of the limited experimental conditions. Based on that, the necessary multiple working conditions were constructed, and the corresponding data were thereafter obtained through simulation of the mechanism model. Finally, the data surrogate mapping model of the thermal system was established based on the data of multiple working conditions, and could be used without a mechanism simulation system. The analysis of a combined heat and power unit shows that the error between the neural network data surrogate mapping model and mechanism model is less than 1%, justifying that the proposed approach has high accuracy.