Development of an artificial neural network model for determination of longitudinal and transverse dispersivities in a convergent flow tracer test

The convergent flow tracer test is an efficient method for determining dispersivity in field, but the traditional curve-fitting method for the estimation of dispersivity from a convergent flow tracer test is quite time-consuming. In this study, we present a model to improve the evaluation of longitudinal and transverse dispersivities from a convergent flow tracer test which couples a back-propagation neural network (BPN) model with a two-dimensional convergent flow tracer transport model. The prediction errors for the training and validation data show that with the effective porosity fitting model, the scale-dependent longitudinal dispersivity fitting model, and the scale-dependent transverse dispersivity fitting model, we can obtain satisfactory prediction accuracy with much less computational time. The applicable ranges of parameters are: The Peclet number is between 0.5 and 100, the effective porosity is between 0.05 and 0.5 and the scale-dependent transverse dispersivity is between 0.01 and 10 m. One set of hypothetical data and one set of field data are used to demonstrate the robustness and accuracy of the back-propagation neural network fitting model (BPNFM). The results demonstrate that BPNFM has the advantage of significantly saving the computational time and giving more accurate transport parameter values. The developed BPNFM is an effective tool for fast and accurate evaluation of the longitudinal and transverse dispersivities for a field convergent flow tracer test.