A hybrid data-model-driven framework for monitoring threshold determination and dynamic updating is proposed, enabling short-term structural response prediction and construction load inversion (back-calculation). The framework consists of three core modules. The first module comprises a visual displacement monitoring system responsible for data acquisition, real-time warning, and continuous information transmission to the second module. The second module determines and updates monitoring thresholds using hybrid data-model-driven methods and provides extensive training samples for the third module. The third module employs the CNN-BiLSTM-AdaBoost algorithm to perform short-term prediction of structural displacement responses and inversion of construction loads, allowing displacement trends to be predicted up to one hour in advance and enabling early warning. The results demonstrate that the proposed monitoring threshold determination and updating method can be continuously refined in engineering practice. The buckling failure displacement predicted by the updated model exceeds the initial alarm threshold, indicating that the initial threshold is conservative and appropriately defined. Moreover, the CNN-BiLSTM-AdaBoost-based approach exhibits strong robustness and high prediction accuracy, enabling real-time response prediction and load inversion, and shows considerable potential for practical engineering monitoring applications.