Abstract: It is the hotpot and challenge in the field of earthquake for charactering seismic intensity attenuation relationship. Many scholars have established some prediction models of seismic intensity attenuation. In order to further improve the prediction accuracy of the seismic intensity attenuation models, 243 isoseismics from 107 earthquake cases are collected from Sichuan-Yunnan to establish a new comprehensive prediction model base on back propagation (BP) neural network. This prediction model fuses the merits of the Western China intensity attenuation relation based on the traditional model and the matrix attenuation relation based on matrix model. The experimental result shows that the prediction accuracy of the proposed fusion model is better than that of the matrix attenuation relationship and the western Chinese intensity attenuation relationship.