Magnetotelluric inversion method based on fully convolutional neural network

US11782183B2 · kind B2 · utility

Assignee

• Institute of Geology and Geophysics, Chinese Academy of Sciences · CN

Inventors

• Zhongxing Wang · Beijing, CN
• Lili Kang · Beijing, CN
• Zhiguo An · Beijing, CN
• Ruo Wang · Beijing, CN
• Xiong Yin · Singapore, SG

Key dates

• Filing date: Jun 14, 2022
• Grant date: Oct 10, 2023
• Expiry date: Jun 14, 2042

Classification

• Technology area (CPC G): Physics
• CPC primary: G06N3/084
• WIPO field: Measurement
• WIPO sector: Instruments

Abstract

Disclosed is a magnetotelluric inversion method based on a fully convolutional neural network. The magnetotelluric inversion method includes: constructing a multi-dimensional geoelectric model; constructing a fully convolutional neural network structure model to obtain initialized fully convolutional neural network model parameters; training and testing the fully convolutional neural network structure model based on the training sets and the test sets to obtain optimized fully convolutional neural network structure model parameters; determining whether training of the fully convolutional neural network structure model is completed according to fitting error changes corresponding to the training sets and the test sets; and finally, inputting measured apparent resistivity into a trained fully convolutional neural network structure model for inversion, and further optimizing the fully convolutional neural network structure model by analyzing precision of an inversion result until an inversion fitting error satisfies a set error requirement.