A New Method for Geophysical Induced Polarization Inversion Based on Stochastic Medium Model and Sample-Compressed Artificial Neural Network

Weiqiang Liu,Liangyong Yang
DOI: https://doi.org/10.1088/1742-6596/2025/1/012010
2021-09-01
Journal of Physics: Conference Series
Abstract:Abstract Induced polarization (IP) is a near-surface geophysical exploration method. Inverting the electrical properties of the underground medium from surface apparent IP parameters is a highly nonlinear problem. To further improve the accuracy, the artificial neural network (ANN) algorithm is applied to the two-dimensional (2D) IP data inversion for the first time. We firstly produced smooth geo-electric models based on the stochastic medium theory, and obtained the corresponding theoretical responses through forward modelling. Then, we compressed the responses and models through image compression technology. Finally, the above compressed responses and models were used as input and output samples to train an optimal network system for inversion. We tested the algorithm with synthetic examples. The results show that ANN can improve the longitudinal resolution of the inversion results and make the inversion results more focused.
What problem does this paper attempt to address?