Data-driven modelling for pore water pressure variation responses to rainfall

Abstract

Measurement of soil pore water pressure is always a tedious, time consuming and expensive exercise. Moreover, unavailability of any physical based or mathematical relationship to get information of pore water pressure leads researchers to perform data-driven modelling. This study presents a data-driven modelling approach to predict soil pore water pressure variations in a slope. Point measurements based time series data of soil pore water pressure variations and corresponding rainfall was used to develop the data-driven model. The model was developed using radial basis function neural network with Multi-quadric basis function. The inputs of the model consist of 5 antecedent pore water pressure, two antecedent rainfall and one current rainfall values. Trial and error procedure was adopted to obtain the appropriate number of neurons in the hidden layer. Normalization method was used to determine the spread of the basis function. Mean absolute error (MAE) and coefficient of determination (R² ) as statistical measures were used to evaluate the performance of the model. The results revealed that the data-driven model predicted the pore water pressure values close to the observed values. The minimum value of MAE during test stage was observed as 0.327 with a coefficient of determination R² = 0.975. Multi-quadric basis function was found to be suitable for the prediction of soil pore water pressure variations.