BIM models have played a central role in the construction industry in recent years and become an active research direction for intelligent buildings and intelligent cities. Room is an important part of BIM model, and the perfection of room data is of great significance to improve the automation, efficiency and accuracy of BIM applications such as compliance inspection and spatial analysis. to the problem that insufficient room information in the BIM model leads to the difficulty of identifying the room, a room function type identification method combining simulated annealing and SVM is proposed and verified on the residential model set. first, based on the Revit platform secondary development technology, the room shape feature parameters are obtained first, and the main parameters of the SVM model are optimized by expanding the spatial syntactic variables such as connectivity, control, average depth, integration, intelligence. finally, the accuracy of the optimized SVM model is verified on the validation set. experimental results show that compared with the algorithms proposed in other similar literatures, the accuracy of the proposed algorithm can reach 95.59%, which is better than that of similar algorithms. experiments on four UCI datasets verify the stability and effectiveness of the classification model.