Real-Time Attitude Prediction and Dynamic Monitoring of Super-Tall Buildings Using TQWT-TCN-LSTM-GAM Integrated with GNSS Multi-Antenna Systems

Abstract. The safe operation of supertall buildings urgently requires real-time, high-precision attitude monitoring. However, under the coupled influence of complex operational loads and extreme environmental factors such as wind, earthquakes, temperature, and humidity, GNSS-based deformation monitoring systems often face challenges such as discontinuous data acquisition, fluctuating accuracy, and strong nonlinearity. These issues hinder the ability to meet the demands for real-time, high-precision, and high-frequency structural health monitoring of supertall buildings. To address these challenges, this paper proposes an attitude monitoring method that integrates Tunable Q-factor Wavelet Transform (TQWT) with a Temporal Convolutional Network–Long Short-Term Memory (TCN-LSTM) hybrid neural network model. The model adopts a serial architecture to significantly enhance the capability of processing attitude information. Additionally, a Global Attention Mechanism (GAM) is incorporated to improve the model’s sensitivity, responsiveness, and representational accuracy for local anomalies and non-stationary features, thereby enabling real-time and high-precision monitoring of torsional deformations in supertall buildings. Based on this approach, a spatiotemporal prediction model for building attitude was developed. Validation experiments using a GNSS multi-antenna system demonstrated that the proposed TQWT-TCN-LSTM-GAM model achieves significantly higher prediction accuracy under complex environmental conditions compared to traditional neural network and machine learning methods.