Settlement monitoring and prediction are among the most concerning problems for an immersed tunnel. However, settlement monitoring usually relies on manual surveys with low frequency, while the settlement prediction for the entire tunnel is usually based on the assumption of an elastic foundation. Therefore, determining the impact of short-term events on the settlement of an immersed tunnel is difficult. This paper combines a physics- informed machine learning algorithm with the Kelvin-Voigt model, introducing a retardation time factor to express the delayed response to an applied force. This framework is further used in the Hong Kong-Zhuhai-Macau Bridge tunnel, and the effect of an extreme water level caused by a typhoon on this tunnel is analyzed.