Prediction of soil thermal conductivity based on Intelligent computing model

Thermal conductivity is a basic characteristic of the heat conduction properties of subsoil. Previous research shows that soil thermal conductivity has complex correlations with many soil physical parameters, such as dry density, water content, mineral composition and particle-size distribution. In this paper, several artificial intelligence calculation methods are used to study the soil heat conduction mechanism and establish predictive models of thermal conductivity: an artificial neural network (ANN), adaptive neural network-based fuzzy inference system (ANFIS) and support vector machine (SVM). Their modelling performance was evaluated by several metrics: correlation coefficient (R2), root mean square error (RMSE), mean absolute error (MAE) and variance account for (VAF). Monte Carlo simulation was used to verify the robustness of the models, and the results of traditional empirical relationship models are used for comparison. The ANN, ANFIS and SVM models can accurately predict soil thermal conductivity, with R2 > 0.89, RMSE < 0.22 (Wm−1 K−1), MAE < 0.14 (Wm−1 K−1) and VAF > 88%. The ANN model had the best predictive accuracy, with R2 = 0.9535, RMSE = 0.1338 (Wm−1 K−1), MAE = 0.0952 (Wm−1 K−1) and VAF = 95.25%. The SVM model had similar accuracy, while that of the ANFIS model was lower. Monte Carlo simulations show that the SVM model provided the most robust predictions and that all three models were significantly better than the traditional empirical models. The SVM model is suggested as the best model for predicting soil thermal conductivity.