Utilization of neural network feedback method to prediction of thermal resistivity of soils

In order to study the heat transfer characteristics of different soils, the correlation between the thermal resistivity of soil and the main influencing factors is analyzed briefly through the literature data. A prediction model for thermal resistivity of soils by the using neural network is proposed, and the effectiveness and superiority of the proposed model are compared. The measured thermal resistivity is compared with the predicted results of the feedback neural network model. The results show that the feedback neural network can accurately and effectively predict the thermal resistivity of soils. The model adopts dry density, saturation and quartz content as the input parameters, which comprehensively and reasonably reflect the main factors affecting the thermal conductivity of soils. The prediction model has high precision, the correlation coefficient R² of the predicted and measured values is greater than 0.93, the root meansquare error (RMSE) is lower than 28 K∙cm/W, and the variance accounting for (VAF) is greater than 94%. Compared with the traditional empirical relationship, the feedback analysis model has significant advantages in the predicted results in the new environment.