Predicting Soil Electrical Resistivity Using Geotechnical Properties and Artificial Neural Networks

This study investigates the influence of key geotechnical parameters—water content, dry density, and plasticity index—on soil electrical resistivity, with the goal of improving prediction accuracy for substation grounding system design. A dataset comprising 150 laboratory test results was compiled from soil samples collected at three substations in Thailand, representing diverse moisture conditions to reflect field variability. Two modeling approaches were applied: multiple regression (MR) and artificial neural networks (ANN), evaluated using the coefficient of determination (R²) and root mean square error (RMSE). The MR models achieved relatively strong correlations, with R² values up to 0.8281; however, their higher RMSE values indicated limited precision under variable conditions. In contrast, the ANN models, particularly those incorporating the plasticity index, demonstrated superior performance, achieving lower RMSE values—down to 0.057—highlighting their ability to capture complex nonlinear relationships. In comparison to prior studies that often relied on single-variable models or uniform soil datasets, this research adopts a more integrative and generalizable framework. By incorporating multiple soil parameters into the ANN model and validating against a diverse dataset, the study offers practical insights for engineering applications. The findings are particularly valuable in tropical regions where soil moisture variation significantly impacts resistivity and grounding system performance.