Deep Learning-Based Electrical Exploration for High-Precision Permafrost Inversion Method

As the acceleration of global warming and humidification continues, the thickness of the active layer in permafrost regions is increasing, and the permafrost table is significantly lowering. The melting of permafrost has triggered a series of engineering problems, such as uneven settlement and deformation of roads, tilting, cracking, and even collapse of buildings. Therefore, accurately detecting the distribution of ice content in the subsurface of permafrost regions is of great significance for the construction of new permafrost projects and the disaster prevention of existing projects. Currently, the detection of ice content in permafrost primarily relies on high-density electrical exploration, which is based on the resistivity differences between soil and ice. However, due to the small electrical differences between permafrost with varying ice content, existing methods can only roughly determine the position of the permafrost upper limit, and it is difficult to accurately determine the thickness of the active layer and the specific ice content of the permafrost. To address this issue, this paper proposes a high-density electrical inversion method based on deep neural networks. By incorporating physical laws into the inversion process, the inversion accuracy of high-density electrical exploration in permafrost areas is significantly improved. In field exploration experiments conducted in the Qinghai-Tibet Plateau, the inversion results of this method were highly consistent with the results of borehole measurements, validating its effectiveness.

Keywords: permafrost, deep  learning,electrical exploration