Turning data into better wells

Posted 16.10.2017 13:16 by Magnus Tvedt

Well Planning is one of the most demanding aspects of drilling engineering. In this aspect of the oil and gas industry, the data is spread across different domains and at different locations. Data analysis and machine learning could be used in this process by aiding drilling engineers to design new wells.

By Khushal Adlakha, Data Scientist at Pro Well Plan

Well Planning is one of the most demanding aspects of drilling engineering. In this aspect of the oil and gas industry, the data is spread across different domains and at different locations. Hence the data is not analysed properly before a detailed well plan is provided to a drilling engineer. Data analysis and machine learning could be used in this process by aiding drilling engineers to design new wells. This post will give an insight on using historical data for making machine learning driven decisions in well planning and operations.

For the base case study, the open dataset from the Norwegian Continental Shelf (NCS) was analysed. Norwegian Petroleum Directorate (NPD) contains aggregate overviews of exploration and development wells on the NCS. The publicly available dataset contains limited attributes of the well. For investigating the above data, Python programming language was used. We chose Python because of its simple and elegant nature. Different python libraries for data mining and machine learning techniques like pandas, matplotlib, seaborn and scikit-learn makes this work interesting and enlightening.

Pandas is one of the finest library for data manipulation and analysis. We can apply different data operations on numeric tables and series data. That means our data needs to be converted into tabular format before we start playing with the data. Matplotlib and seaborn are data visualization libraries which includes cool interfaces for making statistical graphics. Scikit-learn is an efficient tool for machine learning algorithms that can revolutionize automation in oil and gas industry. The best part of python and all the above libraries are that they are easily accessible to everybody.

According to a New York Times study, 80% of the time is invested in cleaning and preparing the data in order to produce better prediction models, and this study was no different. Different data cleaning techniques were applied during this study before we can delve into machine learning. For instance, data types were converted from integer to float for some of the attributes and different mud types were converted into categorical variables etc.

Once the data was cleaned and organized in a proper format, different parameters like mud weight and mud type were then predicted using algorithms like random forest classifier (from scikit-learn) and xgboost. The results were decent enough to continue our study of machine learning driven decisions. The below points are just the example of different workflows where machine learning driven decisions can come handy.

  • Mud type can be predicted by taking different formations, mud weight and mud sample depth point as input parameters.
  • Bit parameters can also be predicted by using depth sections as input. Consequently bit size, bit type, WOB and RPM can also be predicted.

In this way, data analysis and machine learning would be able to integrate all the historical drilling events and will be able to optimize well planning and operation based on the geology and the area.

Next phase is to apply the same concept on the data of a client company.

Similarly, machine learning has got huge potential in automating the operations tasks. Real time information gathering is of utmost importance in conducting safe and efficient drilling operations. Hence, modern rigs have numerous sensors actively measuring various parameters related to the down-hole drilling environment. By comparing and investigating data from hundreds of sensors around the field, different properties like weight on bit, drilling speed etc. could be adjusted in real time and thus leading to reduction in time. The sensors spread across all the wells in the fields will eventually help to drill a new well in half the time as compared to previously drilled wells. One important advantage of using machine learning systems and sensor data is the direct reduction in oilfield data gathering personnel working at rigs.

The applications of machine learning in well planning and operations could save billions of dollars for the operating companies. Digital well planning can produce more consistent results by cutting out human errors. Hence, attention could be given on performance of the well and safety of the crew.

Stay tuned for exciting findings from the client company’s data.

Sign up for our newsletter

Contact

The Team

Pro Well Plan AS is based in Bergen, Oslo and
Stavanger
Magnus Tvedt
CEO Magnus Tvedt magnus.tvedt@prowellplan.com
Nicholas Mowatt Larssen
CTO Nicholas Mowatt Larssen nimola@prowellplan.com
Cathrine Tangerås Eide
Project Manager Cathrine Tangerås Eide cathrine.eide@prowellplan.com
Khushal Adlakha
Data Scientist Khushal Adlakha khushal.adlakha@prowellplan.com
Torgeir Lassen
CFO Torgeir Lassen torgeir.lassen@prowellplan.com
Eirik Lyngvi
Software Developer Eirik Lyngvi eirik.lyngvi@prowellplan.com

Send a message

© 2018 Pro Well Plan AS - All rights reserved