Abstract: Conducting landslide susceptibility assessments in the Winter Olympics region is important for the operation and maintenance risk management of the Winter Olympics venues. In view of the tedious process of repeatedly adjusting the susceptibility factor weights and the large loss of feature information due to too many pooling layers,a Weighted Multi-CNN(WM-CNN) is proposed for landslide susceptibility prediction with adaptive learning of the influence factor weights and the replacement of pooling layers by dilated convolutional layers. This paper takes six districts and counties of the Winter Olympics region as research objects,constructs the evaluation index system of landslide susceptibility in the Winter Olympics area from five aspects: topography and geomorphology,geological structure,hydrology,human activities,and soil vegetation,and uses the WM-CNN model to construct the evaluation model of landslide susceptibility. The ROC curve is used as the accuracy index,and the accuracy is compared with the one-dimensional convolutional neural network(CNN-1D),convolutional neural network(CNN),support vector machine(SVM),and random forest(RF)models. The results show that the WM-CNN model has the best prediction effect,which is higher than 0.835 for the CNN-1D model,0.877 for the CNN model,0.819 for the SVM model,and 0.884 for the RF model. In addition,the very high susceptibility and high susceptibility areas in the study area are concentrated in the Yanqing district of Beijing,mostly on both sides of the road and in the valley. The National Ski Jumping Center and Yanqing Olympic Village are located in the medium susceptibility area with higher landslide risk and thus need to be monitored in a focused manner.