Affiliation:
1. Department of Electrical Engineering, Faculty of Engineering, Shahid Chamran University of Ahvaz, Ahvaz, Iran
2. Department of Electrical Engineering, Faculty of Engineering, Shahid Chamran University of Ahvaz, Ahvaz, Iran (Corresponding author)
3. National Iranian South Oil Company, Ahvaz, Iran
Abstract
Summary
Cementing, as a common operation in well drilling and completions, requires accurate calculations and feasible studies. High-quality well cementing is necessary to achieve objectives and to prevent accidents and damage. The correct levels of cementation and cementation of good quality must be achieved, and thus both of these must be accurately evaluated. Cement bond quality can be assessed in various methods, for instance, the use of acoustic equipment. Acoustic readings can be processed to generate graphics of the cement status. These images are inspected by experts who assess the cement bonding visually, providing qualitative estimates of their quality. Regarding the explanations, industry needs automatic, fast, and intelligent methods. In this study, the performance of neural networks for automatic interpretation was evaluated and combined with fuzzy systems. We present a convolutional neural network (CNN) to address these challenges. The data were extracted from the oilwell logs of District 1 of Ahvaz, sampled by the National Iranian South Oil Company (NISOC), and were divided into three classes of cementing bond quality—namely, good, midrate, and bad. Each class contained 1 m data entries. The input to the CNN comprised images of sampled variable density logs (VDLs). In the second stage of analysis, the weights of the learned networks were optimized and replaced with a Bees algorithm (BA) and a fuzzy system (fuzzy-CNN). The estimated efficiency of the CNN to classify the well cement quality and fuzzy-CNN was 80.15 and 83.62%, respectively. The study provides insight into the efficiency and outputs of each of the networks studied. Problems encountered in training the networks to operate optimally as an automatic interpretation system were investigated and are discussed.
Publisher
Society of Petroleum Engineers (SPE)
Subject
Mechanical Engineering,Energy Engineering and Power Technology
Reference25 articles.
1. Research Predicts Improved Cement Bond Evaluations With Acoustic Logs;Anderson;J Pet Technol,1961
2. Automatic Well Log Analysis Across Priobskoe Field Using Machine Learning Methods;Belozerov,2018
3. Committee Neural Networks for Porosity and Permeability Prediction from Well Logs;Bhatt;Geophys Prospect,2002
4. Evaluation of Deep Learning Models for Multi-Step Ahead Time Series Prediction;Chandra;IEEE Access,2021
5. Understanding Dropout: Training Multi-Layer Perceptrons with Auxiliary Independent Stochastic Neurons;Cho,2013
Cited by
3 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献