Adaptive State-Space Kalman Filter Model for Tunnel Deformation Prediction

Abstract

Accurately predicting tunnel deformation to ensure construction safety, optimize support, and issue structural risk warnings. When settlement gages, convergence gages, inclinometers, and stress sensors are used for monitoring, phased evolution, measurement noise, and incomplete observations often occur. This article proposes a new adaptive state-space Kalman filter model for predicting tunnel deformation. The model uses displacement, deformation velocity, acceleration, surrounding pressure, and construction phase parameters to illustrate the development of deformation. Using the state-space refinement method, update the state transition matrix and noise covariance based on the most recent monitoring residuals. Finally, the optimized Kalman filter calculates the alarm probability and short-term deformation prediction. Based on experimental analysis, ARIMA, support vector regression, neural networks, standard Kalman filters, and deep sequence models were tested. The tunnel monitoring sequence consists of multiple segments. The new technology reduced the noise observation RMSE to 1.36 mm and improved the early warning recall rate to 93.5%. This paper proposes a tunnel deformation monitoring system prediction method that is easier to use and understand.