The fundamental principles of three representative machine learning models-Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory networks (BiLSTM), and the ensemble learning algorithm AdaBoost-are systematically analyzed. Based on the respective advantages and limitations of individual models, a CNN-BiLSTM-AdaBoost hybrid prediction model is developed. Furthermore, an overall framework and implementation pathway for hybrid data-model-driven monitoring of major hazardous engineering structures are proposed, providing methodological and theoretical support for subsequent prediction of tower crane dynamic responses under typhoon loading, dynamic determination and updating of monitoring and warning thresholds for high formwork support structures, and short-term structural response prediction. In this framework, multi-source monitoring data are acquired in real time by various sensing devices and synchronously transmitted to an early warning platform for integration, visualization, and analysis, while simultaneously driving the dynamic updating of finite element models. Based on measured data and simulation results from the updated models, safety evaluation and response prediction are conducted, and the assessment results and warning information are fed back to the construction site, thereby realizing a closed-loop interaction among the physical structure, the finite element model, and the monitoring and early warning platform.