The color of the LED smart light is tunable by its inner equipped micro-processing systems. Therefore, it could provide significant improvement for the smart lighting conditions, such as museum lighting and home lighting. At present, the limitation of the current lighting blending technology remarkably affects the application of smart lighting technology and people could not make full use of the adjustability of the smart luminaries. In this research, a novel light blending model was proposed based on BP neural network and active set algorithm. The models could effectively simulate the nonlinear relationship between the device control values of the smart light and the output radiance spectrum of the light. Particularly, a BP neural network-based forward model for LED light blending was firstly proposed, which could accurately calculate the spectral radiance power distribution from the device control values. Afterwards, based on forward model, an active set algorithm-based backward model was developed, which could precisely predict the device control values from the desired spectral radiance power distribution. The experimental result indicates that the proposed method could accurately achieve the light blending controlling of smart LED light, with a CIEUCS Duv value of 0.002 7, which is significantly smaller than the just noticeable difference value of human vision. The authors believe that the proposed method will provided effective support for the development of smart LED lighting in near future.