Ground Prediction by Markov-ANN Hybrid Analytics Using Baseline Geotechnical Data and Observed Field Data

Abstract

ABSTRACT Geology along the tunnel length usually differs from the anticipated data as per Geotechnical Baseline Report (GBR), sometimes quite significantly. This has always led to the disruption of the planning, resource mobilisation, cost and duration of the project. One way to minimise this uncertainty is by exhaustive investigation like probe holes and geophysical prospecting. Forecasting or prediction of Geology can also be done by soft computing and analytical methods. Using the analytical tool of the Markov Model and the learning powers of Artificial Neural Networks (ANN) has been attempted in this study. The idea is to build a hybrid model that would combine both capabilities and do a probabilistic prediction of the ground condition ahead of the tunnel face in real time to better suit the site, accommodate complexities and can capture the associated uncertainties. The Geotechnical Baseline Report (GBR) having the initial survey details of the tunnel geology is used to build the Markov Model while the deterministic borehole data in the GBR is used to build the ANN model. Bayesian joint probability theorem is used to update the model with the face observed geological parameters. This hybrid model demonstrated to be a good complement to the physical forecasting methods and the geological uncertainty of a complex and challenging Himalayan region was well captured by the present approach. The window of the prediction was for a region of approximately 250m where it showed a very good prediction of ground class. The model can help in systematic planning and resource mobilisation in a better way and subsequent key decisions for the upcoming excavation method and corresponding support measures.