Very soft organic clay applied for road embankment
It is important to pre-process the data to a suitable form before they are applied to the ANN. Pre-processing the data by scaling them is important to ensure that all variables receive equal attention during training. The output variable has to be scaled to be commensurate commensurate with the limits of the transfer functions used in the output layer. In this study, the input and output variables are scaled between 0 – 1. 5.3.4 Model Architecture
Determination of the network architecture is important in the development of ANN models. It requires the selection of the number of hidden layers and the number of nodes in each of these. It has been shown that a network with one hidden layer can approximate any continuous function, provided that sufficient connection weights are used (Hornik et al., 1989). Consequently, one hidden layer is used in this study. The number of nodes in the input and output layers are restricted by the number of model inputs and outputs. Table 5.2 shows the summary of input and output layers of each ANN model. In order to obtain the optimum number of hidden layer nodes for each model, each ANN model is trained with one to ten hidden layer nodes. The smallest network that is able to map the desired relationship is used. The coefficient of 2 determination (r ), the root-mean-square error (RMSE), and the mean absolute error (MAE) are the criteria used to evaluate the performance of the ANN models. The results of the evaluations are shown in Appendix G. The smallest number of hidden 2 layer nodes, with the highest value of r and the lowest values of RMSE and MEA, is selected as optimum. The evaluation resulted in an optimum number of hidden layer nodes for the ANN models for estimating the settlement, settlement time, and embankment stability are six, eight, and eight nodes, respectively. Table 5.2. Summary of input and output layers of the ANN models Input layer ANN model
Settlement
Time of settlement
Embankme nt stability
Num. of nodes 4
5
3
Model inputs Plasticity index (PI) Initial void ratio (eo) Layer thickness (C) Applied load (L) Plasticity index (PI) Initial void ratio (eo) Layer thickness (C) Applied load (L) Settlement (SET) Undrained shear strength (Cu) Layer thickness (C) Embankment height (He)
Hidden Output layer layer Num. of Num. of Model nodes nodes outputs 6
8
8
1
Settlement (SET)
1
Settlement time (TIME)
1
Max. surcharge load (S)
5.3.5 Weight optimisation (training)
The process of optimising the connection weights is known as training. A feed forward network and a back-propagation algorithm, commonly used for finding the optimum weight combination, are used in the ANN models. The feed forward networks, trained with the back-propagation algorithm, have already been applied successfully to many
