Research on building construction process monitoring and intelligent prediction model based on deep learning

This study focuses on the monitoring and intelligent prediction of building construction process, and proposes a model framework based on deep learning. According to the characteristics of complex and multi-source data in construction site, a multi-modal data collaborative processing framework is constructed, which integrates monitoring video, sensor data and construction logs. In the aspect of anomaly detection, the improved YOLOv5s+Transformer architecture is adopted to realize the efficient identification of construction safety hazards, and its [email protected] reaches 89.7%, FNR is 5.2%, and the delay is only 38ms, which is significantly better than the traditional method. For the prediction of construction progress, the spatio-temporal graph convolutional networks (ST-GCN) model is used to effectively capture the information of time and space dimensions. In the 7-day, 14-day and 30-day prediction periods, the RMSE is 0.89, 1.23 and 1.47 respectively, which shows high prediction accuracy. In the quality prediction module, physical information neural network (PINN) is introduced, which combines physical principles with neural network training to greatly reduce the prediction error of concrete strength. Through multi-modal data fusion, the performance of the model in safety monitoring, progress prediction and quality prediction is further improved. The experimental results show that the model has wide adaptability and high accuracy in various construction scenarios, and it can still maintain good performance in extreme weather conditions and data fusion of different brands of sensors. In cross-site migration learning, domain adaptation and federated learning methods can effectively improve the performance of the model in the new site environment, although it requires a certain training cost, but it has high application value.