PREDICTION OF HYDRAULIC CONDUCTIVITY OF CLAY LINERS USING ARTIFICIAL NEURAL NETWORK

Abstract

This paper pertains to prediction of hydraulic conductivity of soil used as clay liners using artificial neural networks based on soil classification test results like Atterberg’s limit, grain size and compaction characteristics. Feed forward back propagation neural network has been used and trained with different combination of input parameters of laboratory test results available in literature. Statistical performances criteria like root mean square error, correlation coefficient, coefficient of determination and overfitting ratio are used to compare different neural network models, the available statistical model and the results obtained using group method of data handling (GMDH) neural network. The neural network models are found to be more efficient and reliable compared to statistical model. Identification of important soil parameters affecting the hydraulic conductivity of soils is discussed. A model equation is presented with weights of the trained neural network as model parameter.