Continuous Prediction of Creep in a Slope Utilizing Inclinometer Data

Creep of natural slopes is difficult to quantify, especially with regard to variations due to seasonal changes. A quick clay slope in Norway was instrumented with 71 inclinometers. The inclinometers were also meant to monitor the influence of construction works taking place in the area. Distinguishing deformations due to creep and deformations due to construction works can be complexespecially because ofseasonal variations. Inclinometerswere installed from 6 to 35 months before the construction works started, and results were used to develop a method for separating the influence of construction and creep on the measured deformation.Various types of prediction models were evaluated, from different curve fitting methods to more advanced machine learning models. A random forest machine learning algorithm utilizing meteorological data and historical measurements proved to give the most reliable prediction, reducing the mean absolute error of the predicted deformation by more than 50% compared to the conventional curve fitting methods.

Keywords: Geotechnical engineering, Creep, Deformation, Inclinometers, Machine learning, Prediction model.