z-logo
Premium
Calculation of the thermal conductivities of fine‐textured soils based on multiple linear regression and artificial neural networks
Author(s) -
Wen Haiyan,
Bi Jun,
Guo Ding
Publication year - 2020
Publication title -
european journal of soil science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.244
H-Index - 111
eISSN - 1365-2389
pISSN - 1351-0754
DOI - 10.1111/ejss.12934
Subject(s) - thermal conductivity , mean squared error , saturation (graph theory) , linear regression , dryness , artificial neural network , soil water , soil science , water saturation , mathematics , biological system , statistics , materials science , environmental science , machine learning , computer science , composite material , porosity , biology , medicine , surgery , combinatorics
Abstract The thermal conductivity of soils is an important parameter in environment, earth science and engineering applications. A new thermal conductivity model based on the Fredlund and Xing model was proposed to mimic the thermal conductivity of fine‐textured soils from dryness to full saturation. However, the use of a measured thermal conductivity value at a certain saturation to calculate the parameter of the thermal conductivity model reduces the ease of use. Therefore, the objective of this study was to use multiple linear regression (MLR) and artificial neural networks (ANN) to improve the thermal conductivity model. The performances of the MLR‐based model and ANN‐based model were evaluated and compared with measured data and three existing empirical models at different saturation ranges. The results showed that the MLR‐based model and ANN‐based model performed best at low saturations, the ANN‐based model performed best at saturations between 0.01 and 0.6, and the Johansen model showed the best fit to the measured data at saturations of above 0.6. In general, among the models, the ANN‐based model performed best from dryness to full saturation (root mean square error (RMSE) = 0.098 Wm −1 K −1 and average absolute deviation (AAD) = 0.071 Wm −1 K −1 ), followed by the MLR‐based model (RMSE = 0.110 Wm −1 K −1 and AAD = 0.081 Wm −1 K −1 ). The MLR‐based model and ANN‐based model are promising for the accurate calculation of the thermal conductivity of fine‐textured soils from dryness to full saturation and can be incorporated into the numerical modelling of heat and mass transfers. Highlights The ANN‐based model performed best among five models from dryness to full saturation, followed by the MLR‐based model. The MLR‐based model and ANN‐based model showed the best fit to the measured data at low saturations. The ANN‐based model showed the best fit to the measured data at saturations between 0.01 and 0.6. The Johansen (1975) model showed the best fit to the measured data at saturations of above 0.6.

This content is not available in your region!

Continue researching here.

Having issues? You can contact us here