A method of crowdsourcing the labeling and evaluating oilfield data
Publication Date: 2016-Dec-12
The IP.com Prior Art Database
A method of crowdsourcing the labeling and evaluating oilfield data, comprising designing a proper database of data and workers, assigning tasks to various workers, and collecting and analyzing results from workers to update a decision making system.
A method of crowdsourcing the labelling and evaluating oilfield data, comprising designing a proper database of data and workers, assigning tasks to various workers, and collecting and analyzing results from workers to update a decision making system.
Artificial intelligence (AI), as well as machine learning (ML), is being and going to be applied more and more widely in oilfield industry, to learn from the existing knowledges on the fields, technologies, and decisions made historically, thus to make better decision accordingly.
As an important workflow in both AI and ML, we need to train a model with certain amount of historic data. Cleaning and labelling such data, especially field data may take unexpected amount of manual work by the domain experts. Also, evaluating the prediction that an AI or ML made is another important step to continuously improve the trained model, which is called self-learning, or reinforced learning. Evaluating such results may need collecting and processing various opinions from domain experts.
This invention provides a system to execute both these tasks (cleaning and labeling data, and evaluating results), using the so called crowdsourcing methodology.
This system does very well match Plan-Execute-Evaluate process in oilfield industry. It will help analyze the gap between plan and execution, thus improve future plan. This could be another application of this invention
· What are the anticipated products and/or markets for this invention?
o To help continuously improve the text mining model is going to used to extract events/activities from drilling reports.
o Other well planning tools, etc., to generate a plan based on the analysis of plan and execution database.
· In operation, how does the invention interact with the upstream and downstream system?
Input would be domain experts’ answers to some questionnaire, output would be ranking & labelling of corresponding properties in a database. For example, what event does a field engineer think happened according to a piece of recorded comment? Another example for evaluation is, based on our prediction of whether one event had happened from one recorded comment, we invite various drilling engineer to evaluate its accuracy.
Such answers would be employed to improve the train model for AL or ML, fo...