An example of predicting bulk density (RHOB) with Keras and illustrating impacts of normalisation on prediction results
Large amounts of information are acquired each day from wells around the globe. Nevertheless, the standard of that data can vary significantly from missing data to data impacted by sensor failure and borehole conditions. This may have knock-on consequences on other parts of a subsurface project, akin to delays and inaccurate assumptions and conclusions.
As missing data is one of the vital common issues we face with well log data quality, quite a few methods and techniques have been developed to estimate values and fill within the gaps. This includes the applying of machine learning technology — which has increased in popularity over the past few many years with libraries akin to TensorFlow and PyTorch.
On this tutorial, we will likely be using Keras, which is a high-level neural networks API that runs on top of TensorFlow. We’ll use it for example the technique of constructing a machine-learning model to permit predictions of bulk density (RHOB). This can be a commonly acquired logging measurement, nonetheless, it could actually be significantly impacted by bad hole conditions or, in some cases, tools can fail, leading to no measurements over key intervals.
We’ll start with a quite simple model, that doesn’t account for normalising the inputs, a typical step within the machine learning workflow. Then, we are going to then construct a second model with normalised inputs and illustrate its impact on the ultimate prediction result.
Step one on this tutorial is to import the libraries we will likely be working with.
For this tutorial, we’d like 4 libraries:
These are imported as follows:
import pandas as pd
from…