Application of Artificial Neural Networks to Predict Variation of Slope Water Table Induced by Rainfall

Abstract

Ground-water levels are a principal factor of slope instability and of significance in other geotechnical engineering problems In this study we propose a backpropagation neural network with 10-fold cross validation method for predicting the ground-water level induced by rainfall the site was selected at South Cross-Island Highway 52K and use the value of monitoring system with ground-water level soil water content and rainfall measurement Furthermore correlation analysis is used to find the potential input variables range for the predicting model such as Pearson correlation coefficient and partial auto-correlation are applied In this study we choose four factor as the model input such as hourly time series of rainfall and borehole water-level depth of unsaturated zone and unsaturated zone moisture content via 4 different algorithms like the one step algorithm the scaled conjugate gradient method the Bayesian algorithm and the Levenberg-Marquardt method to find the best combination for predicting future 12 hours ground-water level It was found that to obtain satisfactory results the network using 4 hours of antecedent precipitation 3 hours of antecedent borehole water-level the time unsaturated zone thickness and turning 48 hours effective accumulated precipitation into unsaturated zone moisture content performed well in simulating the future 12 hours water-level

Date of Award	2020
Original language	English
Supervisor	Wen-Jong Chang (Supervisor)