Unsupervised machine learning technique for classifying production zones in unconventional reservoirs

Abstract

Significant amounts of information are rapidly increasing in bulk as a consequence of the rapid development of unconventional tight reservoirs. The geomechanical and petrophysical characteristics of the wellbore rocks influence the sweet and non-sweet areas of tight unconventional reservoirs. Using standard approaches, such as data from cores and commercial software, it is difficult and costly to locate productive zones. Furthermore, it is difficult to apply these techniques to wells that do not have cores. This study presents a less costly way for the systematic and objective detection of productive and non-productive zones via well-log data using clustering unsupervised and supervised machine learning algorithms. The method of cluster analysis has been used in order to classify the productive and non-productive reservoir rock groups in the tight reservoir. This was accomplished by assessing the variability of the reservoir characteristics data that are forecasted by looking at the dimensions of the well logs. The Support vector machine as a supervised machine learning algorithm is then used to evaluate the classification accuracy of the unsupervised algorithms based on the clustering labels. The application made use of approximately ten different variables of rock characteristics including zonal depth, effective porosity, permeability, shale volume, water saturation, total organic carbon, young's modulus, Poisson's ratio, brittleness index, and pore size. The findings show that both clustering techniques identified the sweet areas with high accuracy and were less time-consuming.

Keywords

• Machine learning
• Sweet spots
• Unsupervised classification
• Supervised classification
• Unconventional reservoirs
• Clustering analysis