Effect of Antecedent Conditions on Prediction of Pore-Water Pressure Using Artificial Neural Networks

The effect of antecedent conditions on the prediction of soil pore-water pressure (PWP) using Artificial Neural
Network (ANN) was evaluated using a multilayer feed forward (MLFF) type ANN model. The Scaled Conjugate Gradient (SCG) training algorithm was used for training the ANN. Time series data of rainfall and PWP was used for training and testing the ANN model. In the training stage, time series of rainfall was used as input data in one model whereas, rainfall and pore water pressure with some antecedent conditions was used in second model and corresponding time series of PWP was used as the target output. In the testing stage, data from a different time period was used as input and the corresponding time series of pore-water pressure was predicted. The performance of the model was evaluated using statistical measures of root mean square error (RMSE) and coefficient of determination (R2). The results of the model prediction revealed that when antecedent conditions (past rainfall and past pore-water pressures) are included in the model input data, the prediction accuracy improves significantly.