2019 Rice Data Science Conference

​WATCH VIDEO​​​

We use Monte Carlo simulation to capture the details of the oil and gas production process in order to forecast the US oil and gas supply. This differs from a pure machine learning model where the process is treated as a black box, and the underlying mechanism is not known. The model utilizes rig count and historical well production data to forecast oil and gas production within the United States. The model first computes the number of first production wells from the rig count by analyzing the drilling process. Then, it forecasts the future production of each well by fitting Arps’ equation. The model forecasts production at the basin level for both tight and conventional wells for a given scenario. A scenario is a set of parameters including future rig count, drill time, completion time, idle time, backlog probability, and initial production.

​WATCH VIDEO​​​

Recently developed machine learning techniques, in association with the Internet of Things (IoT) allow for the implementation of a method of increasing oil production from heavy-oil wells. In contrast to traditional simulations of steam flood, a widely used enhanced oil recovery technique based on principles of classic physics, we introduce here an approach using cutting-edge machine learning techniques that have the potential to provide a better way to describe the performance of steam flood. We propose a workflow to address a category of time-series data that can be analyzed with supervised machine learning algorithms and IoT. We demonstrate the effectiveness of the technique for forecasting oil production in steam flood scenarios. Moreover, we build an optimization system that recommends an optimal steam allocation plan, and show that it leads to a 3% improvement in oil production. We develop a minimum viable product on a cloud platform.

​WATCH VIDEO​​​

Historically, the real-time hydraulic fracturing analytics system (Real-Time Completion system, RTC) has relied heavily on manual labeled data. The manual tasks, including fracture stage start/end labeling and ball pumpdown/seat event labeling, suffer from human bias and inconsistent errors, and can easily take days to review and correct. This paper provides the development and technical details of the automated stage-wise KPI report generator that fills the manual task gaps and provides industry-leading performance. The generator is constructed with two machine learning models that detect the stage start and end and identify the ball pumpdown and seat operations. These tasks are performed based on the reliably available measurements of slurry rate and wellhead pressure, which enable the real-time automated stage-wise KPI analysis, and they also lay the foundation for further advanced analysis regarding real time hydraulic fracture operational decision making.

​WATCH VIDEO​​​

Depth matching of multiple logging curves is essential to any well evaluation or reservoir characterization and can be applied to various measurements of a single or multiple logging curves from multiple wells within the same field. As many drilling advisory projects have been launched to digitalize the well log analysis, accurate depth matching becomes an important factor in improving well evaluation, production, and recovery. It is a challenge, though, due to the unpredictable structure of the geological formations. We conduct a study on the alignment of multiple gamma-ray well logs by using machine learning techniques. The objective is to automate the depth matching task with minimum human intervention. A novel multitask learning approach is presented to optimize the depth matching strategy that correlates gamma-ray logs. The proposed approach can be extended to other applications as well, such as automatic formation top labeling for an ongoing well given a reference well.

Oil and gas companies are used to analyze different types of 2D images. For example, to date rocks, many microfossils are identified and counted on large microscope images. This research is applied to microfossil detection and could benefit the other use-cases.

To automate this cumbersome work, we suggest a hybrid system consisting of two steps: 1) a heuristic over-segmentation to localize regions of interests (ROIs) with traditional computer vision; 2) a Convolutional Neural Network (CNN) trained to classify ROIs. This hybrid system presents two advantages compared to the state-of-art approach of object detection like those applied to IMAGENET. First, data management of supervised CNN classifiers is more flexible because they are trained on ROIs and not on the overall input image. Second, researchers have focused more on CNN classifiers because of their simplicity. 

Finally, we study the quality of the detection of this system on our micro-fossils detection application.

The goal of this work is to demonstrate the detection and extraction of salt tops from seismic data via the application of deep learning. Motivations for automated salt top extraction is a growing necessity in the field of petroleum exploration. Synthetic data is used for automatic label generation and training of a convolutional neural network is capable of predict with higher accuracy the salt top in unseen data during the training. Several experiments were performed and evaluated for exploring the effects of changing various parameters during training. The best model produced in this study provides excellent results when is compared with the interpretation.