Application of Data Analytics to Chemometric Analysis of Conventionally Produced Oil Samples From the Delaware Basin

Abstract

In the digital age, leveraging the enormous amounts of data available is critical to understanding the petroleum system. One example is chemometric analysis (the application of multivariate statistics to the analysis of chemical data), which has been applied to the Permian Basin as a whole. However, previously published studies have lacked data from the Delaware Basin. The standard chemometric methodology is principal component analysis (PCA) and hierarchical clustering (HCA). The objective of this study is to apply Uniform Manifold Approximation and Projection (UMAP) and mean-shift methodology as well as the traditional methodology to the same proprietary dataset from the Delaware Basin and compare their performances to gain new insights into the petroleum genetic groups in the basin. Thirty-three geochemical parameters were selected from the dataset and were given to both algorithms for analysis. UMAP identified seven different groups while PCA and HCA identified four different groups. The groups generated by both models were compared to source rock extracts and oil samples from the Midland Basin that were published by Echegu (2013) for external validation. The comparison revealed that a first order control on the characteristics of the oil families is whether their source rock was deposited in the Permian or Tobosa Basin. Furthermore, PCA and HCA created one group for oils originating from the Tobosa Basin whereas UMAP differentiates the Tobosa Basin oils into two groups. These two oil groups match the geochemical characteristics of the Woodford and Barnett Formations’ source rock extracts from the dataset in Echegu (2013). For the Permian Basin petroleum systems this study shows that the most probable source rocks for the groups produced by both models are carbonate and shale sources in the Bone Spring and Wolfcamp Formations respectively. Furthermore, the differences between the results from both models are not contradictory but are complimentary. The data analytics supports that these differences are the result of UMAP & mean-shift emphasizing geochemical parameters that are related to thermal maturity whereas PCA & HCA emphasizes the depositional environment.