Automated Digital Well Log Aliasing in Python скачать в хорошем качестве

Automated Digital Well Log Aliasing in Python

Destiny Dong, Jesse Pisel, and Michael Pyrcz from the The University of Texas at Austin present on log aliasing! Geologists working with digital well log files spend lots of their time manually aliasing well-log mnemonics to improve readability and consistency between wells. In this study, we create a Python package called AlasKA to automate the process of aliasing well-log mnemonics. This package standardizes well-log mnemonics so that non-experts can use well logs without time consuming data wrangling and munging. Automatic aliasing of mnemonics in AlasKA allows for basin-wide correlations across thousands of wells. This not only permits uniform aliasing standards for all mnemonics, but also offers a sketch of the input data via the heatmap function in the package. The package comprises three parsers: a dictionary with 1,236 mnemonics and labels, a m-ary keyword extractor tree, and a pointer generator recurrent neural network. The dictionary aliases 60% of mnemonics to their labels by looking for exact matches of mnemonics in the dictionary. The m-ary keyword extractor tree takes care of another 20% of the work. It looks for keywords in the description of a mnemonic and traces keywords down from the nodes to find a label. Lastly, the pointer generator utilizes a Seq2Seq model to generate summaries. In this case the model generates the labels, from the mnemonics’ descriptions. In this talk, we will discuss the mechanics of how the package works, how to customize the dictionary, and how to train your own models to make predictions. We will also walk through how to alias a single well and also how to alias all well-log files in a directory.