Digital construction of geophysical well logging curves using the LSTM deep-learning network

Guozhong Gao

Abstract:

A complete well logging suite is needed frequently, but it is either unavailable or
has missing parts. The mudstone section is prone to wellbore collapse, which
often causes distortion in well logs. In many cases, well logging curves are never
measured, yet are needed for petrophysical or other analyses. Re-logging is
expensive and difficult to achieve, while manual construction of the missing well
logging curves is costly and low in accuracy. The rapid technical evolution of
deep-learning algorithms makes it possible to realize the digital construction of
missing well logging curves with high precision in an automated fashion. In this
article, a workflow is proposed for the digital construction of well logging curves
based on the long short-term memory (LSTM) network. The LSTM network is
chosen because it has the advantage of avoiding the vanishing gradient problem
that exists in traditional recurrent neural networks (RNNs). Additionally, it can
process sequential data. When it is used in the construction of missing well
logging curves, it not only considers the relationship between each logging
curve but also the influence of the data from a previous depth on data at the
following depth. This influence is validated by exercises constructing acoustic,
neutron porosity, and resistivity logging curves using the LSTM network, which
effectively achieves high-precision construction of these missing curves. These
exercises show that the LSTM network is highly superior to the RNN in the digital
construction of well logging curves, in terms of accuracy, efficiency, and
reliability