Heat Flow Prediction in Songliao Basin Based on Deep Neural Network

Abstract

The heat flow is the key data for accurately predicting the contribution of underground heat, but due to the high cost of measurement, the spatial difference is huge, and the heat flow information in some areas is little known. This paper uses deep neural network technology to perform machine learning on a large number of relevant geological and geophysical features and heat flow measurements on a global scale. In addition, the model uncertainty quantification process is introduced, and the reliability of the prediction results is discussed based on the global sensitivity analysis and correlation calculation of the deep learning method. Finally, for the distribution map of heat flow and absolute difference in the Songliao Basin in the study area, the measured data error is less than 11%, and the predicted absolute difference is less than ±20 mW/m2. The results prove the accuracy of the method and the abnormal high heat flow in the central region of the basin. The existence of, further combined with the results of geophysical surveys, provides evidence for the analysis of regional thermal causes.