Big Data Modeling and Intelligent Management Algorithms for Cultural Heritage under Digital Twin Technology

Abstract

Against the backdrop of accelerated globalization and urbanization, cultural heritage faces multiple threats from environmental erosion, human intervention, and natural disasters. Traditional manual inspection methods can no longer meet the needs of real-time and precise preventive protection. This study proposes an integrated "sensing-construction-computation-management" technical route to construct a digital twin framework based on the fusion of PointNet++ and LSTM to achieve synchronous mapping of millimeter-level geometry and minute-level environmental response. Arrays of Faro lidar and DJI oblique imagery were deployed in Dunhuang Mogao Grottoes, Yingxian Wooden Pagoda, and Liangzhu Ancient City to acquire 273 days of continuous data. Geometric features were extracted using four layers of PointNet++ and fused with a two-layer LSTM to obtain a 20-dimensional environmental time series. After FC dimensionality reduction, the health status was output. Sliding window incremental updates combined with EWC regularization achieved an online evolution of 18.7 seconds. CIDOC-CRM knowledge graph-driven Bayesian network hierarchical early warning was implemented. Experimental results show that the system's end-to-end latency of 18.9s is only 15.8% of the golden 120s regulatory limit, the early warning accuracy reaches 96.8%, the macro average F1 is 95.06%, and the performance degradation under extreme environmental disturbances is <4%, which is significantly better than PointNet, DGCNN and LSTM-only baselines. This study achieves second-level response in cultural heritage scenarios, providing a feasible and scalable edge intelligence paradigm for preventive protection.