Artificial neural network for predicting the thermal conductivity of soils based on a systematic database. (July 2022)

• Record Type: Journal Article
• Authors: Li, Kai-Qi; Kang, Qing; Nie, Jia-Yan; Huang, Xian-Wen
• Is Part Of: Geothermics. Volume 103(2022)
• Publication Date: 2022-07
• Subjects: Machine learning -- Artificial neural network -- Database of thermal conductivity -- Spearman correlation coefficient -- Prediction model; Hydrogeology -- Periodicals; Geothermal resources -- Periodicals; Énergie géothermique -- Périodiques; GEOTHERMAL ENGINEERING; GEOTHERMAL ENERGY; GEOTHERMAL EXPLORATION; Geothermal resources; Hydrogeology; Periodicals; Electronic journals; 621.44
• DOI: 10.1016/j.geothermics.2022.102416
• Languages: English
• ISSNs: 0375-6505

Abstract

Highlights: A systematic database of soil thermal conductivity is established. Spearman correlation coefficient ranks the importance of influencing factors. ANN model with remarkable performance to access the thermal conductivity is proposed. Abstract: Thermal conductivity is a significant soil property that affects subsurface temperature distribution and plays an essential role in geotechnical engineering. Accurate evaluation of thermal conductivity is a challenging task since it can be affected by many factors. Although many prediction methods for thermal conductivity have been proposed, most of them are derived from limited experimental results and applied to specific soils. A unified prediction model that is suitable for various soils is still unavailable. This study establishes a general database including thermal conductivity and corresponding physical parameters obtained from experimental measurements and field tests. The Spearman correlation coefficient ranks the importance of influencing factors. In addition, a typical artificial neural network model is conducted and trained by 2197 samples to predict the thermal conductivity of soils. Results show that the thermal conductivity of soils has strong correlations with saturation, porosity and density. The proposed machine learning model has remarkable performance compared to existing prediction models for estimating soil thermal conductivity. This study builds a relatively systematic database of soil thermal conductivity and establishes a unified model to access the thermal conductivity of soils.