Reliable simulation analysis for high-temperature inrush water hazard based on the digital twin model of tunnel geological environment

In complex mountainous terrains, tunnel construction faces unique challenges from high-temperature water inrush hazards, a systemic risk arising from the interplay of stress, seepage, and temperature fields . Traditional simulation methods, focusing on isolated disaster scenarios, fall short in addressing the multifaceted nature of these risks due to geological ambiguity and data incompleteness. Digital twin technology presents an effective solution to these challenges; however, its core challenge lies in how to utilize digital twin technology for data-model-co-driven simulation analysis of coupled multi-physical fields in situations of incomplete data. This paper introduces a digital twin paradigm for the simulation analysis of water inrush, which significantly enhances efficiency and accuracy through the integration of advanced machine learning and finite element analysis techniques. Specifically, this is achieved by combining a high-precision geological modeling method based on Gaussian Processes (GP) with a parameter calibration method through Gaussian Process-Differential Evolution (GP-DE) back-analysis. Firstly, a voxel structure is utilized to integrate the multi-field attribute features of the tunnel environment. Secondly, through the integration of multi-source advanced geological prediction data, we construct a dynamic digital twin model of the tunnel environment leveraging machine learning techniques. To overcome the issue of low modeling accuracy, the GP is employed, enhancing the exploitation of latent information within multi-source geophysical data. Lastly, we utilize the GP-DE back-analysis method to calibrate the parameters of the tunnel environment, thereby enhancing the precision and reliability of water inrush simulations. The method has been validated through application to a section of an ultra-high-temperature water inrush tunnel in China, featuring a burial depth of 230 meters. The accuracy of the method is corroborated by the monitoring data from the tunnel, supporting dynamic optimization design and safety prevention measures during construction. • Introduces a digital twin for simulating high-temperature water inrush in tunnels. • Enhances geological modeling by integrating multi-source data and machine learning. • Calibrates tunnel parameters efficiently using the GP-GE back-analysis with a THM model. • The method was validated in a ultra-high-temperature water inrush tunnel in China.