The Use of Neural Networks with Backpropagation of Error in the Problems of Oil and Gas Well Electrometry

Abstract

Introduction. The final step of electrometry (the main method of geophysical investigation of wells) is quantitative interpretation. Such an interpretation requires solving the ill-posed inverse problem of determining the geoelectrical parameters of the stratification of layers penetrated by the well.
Problem Statement. The need to solve inverse mathematical problems of electrometry of oil and gas wells
presents the challenge of their instability. For electrical logging problems, there is no universal regularization
method for effectively solving ill-posed inverse problems; for induction logging, the development of regularization
methods is a technically complex task.
Materials and Methods. To solve the problem, various parameters and architectures of the neural network
have been tested. A two-layer network with backpropagation of error has been selected.
Purpose. To demonstrate the possibility of effectively solving the inverse problem of electrometry (for both electrical and induction logging methods) using neural networks with backpropagation of error and a simple architecture.
Results. A neural network has been developed and trained (including the design of its structure and the computation of the corresponding training arrays) to determine the parameters of a three-layer formation penetrated
by the well.
Conclusions. It has been shown that the problem of determining the radial (along the layer for vertical wells)
distribution of resistivity can be effectively solved using neural networks with backpropagation of error and a simple architecture.