This paper presents a novel technique for planning and execution of Wireline Formation Testing (WFT) jobs using recent applications of machine learning. WFT measurements provide a link between the static petrophysical measurements and dynamic rock-fluid properties for enhanced formation evaluation. However, despite the availability of newer generation tools for these services, there are still obstacles related to formation test job design and real-time optimization. Skilled geoscientists traditionally use quicklook analysis and theoretical models in designing WFT jobs to acquire quality dataset. In practice, operators have access to a detailed data repository. However, the learnings from historical jobs may be overlooked during dynamic real time operations and engineers may typically regress to simpler or experienced based models. Some of the methods are self-made to reduce time and energy in data acquisition. Digitization and automation of data provide an opportunity for application of data analytics to support these real-time measurements for an objective evaluation of the dynamic data.
In this paper we propose use of unsupervised machine learning models, by which routinely recorded real-time measurements can be used to optimize the test design and real-time operations. Basic principles of interpretation of WFT are used to develop quality parameters and classification based unsupervised machine learning models (Self organized maps, K-means and Hierarchical Clustering). These machine based recommendations can aid in designing and optimizing the real-time job while ensuring quality of the results. In this way, highest quality testing results can be acquired to improve the integration of the dynamic data into the petrophysical analysis. This also will enable standards to be established for real-time data acquisition that can save testing time while improving data and quality.
Results show that machine learning models can be a powerful tool to develop a machine based recommendation systems for classification of WFT responses that are dependent on petrophysical parameters.