Integrated TOC prediction and source rock characterization using machine learning, well logs and geochemical analysis: Case study from the Jurassic source rocks in Shams Field, NW Desert, Egypt

Abstract

The machine learning methods and well log mathematical models have been used for predicting total organic carbon (TOC) in Jurassic source rock formations in Northwestern Desert, Egypt. Conventional well log data from two wells have been utilized for source rock study of Jurassic source rocks from Khatatba and Ras Qattara formations in the study area. The source rock is first studied based on geochemical parameters, which include assessments on the type and amount of kerogen present within the source rock samples. The Jurassic source rock samples have great generative potential and consist of mixed kerogen type III and kerogen type II-III. TOC content reaches up to 46.90% for Khatatba and 16.80% Ras Qattara. In the second part of this research, we attempt to characterize the Jurassic source rocks by using mathematical well log models and machine learning methods. GR, RHOB and NPHI well log data were used for TOC prediction using both methods. The quantified TOC results show that the R2 values of well log models are above 0.9 for both formations, whereas the machine learning method using Artificial Neural Network showed R2 value of 0.4. The results from the well log models suggest that they are applicable in the study area. This study has proven that well log data can be used with confidence to evaluate organic source quantity of Jurassic rocks in Northwestern Desert in the absence of geochemical data.