Understanding live oil composition effect on asphaltene precipitation as a function of temperature change during depressurization using machine learning techniques

The study aims to determine the live crude oil compositional feature’s effect on asphaltene precipitation as a function of temperature. In this study, we have applied different modern feature engineering techniques incorporated with machine learning to understand the importance of governing composition features affecting the asphaltene precipitation as a function of temperature during depressurization. To achieve this purpose, different feature selection techniques integrated with the famous random forest (RF) algorithm were applied to the high pressure high temperature (HPTP) experimental data of ten live crude oil samples available in the published literature having outcome as asphaltene precipitation increase or decrease as a result of temperature rise. All data were visualized by using different techniques. Since the data was scarce in the literature, therefore, to avoid overfitting issues the recursive feature elimination with a fourfold cross-validation technique was applied. Random forest algorithm was trained on 60% of the dataset, while testing was done on the remaining 40% dataset. An accuracy of 100% was achieved during the training phase, while it decreased to zero when applied to the testing dataset. The results were validated using a gradient boosting machine (GBM) and found to be in excellent agreement. However, the implementation of other advanced data science techniques aided in improving the accuracy of the testing phase but to very little margin, i.e., from 0 to 25%. Generally, Heavy ends, Light ends and API were found to be the important features in deciding the trend of asphaltene precipitation with temperature changes. Crude oils with higher Heavy ends or decreased API were found to increase asphaltene precipitation when temperature rises. Since, due to the complex relationship of asphaltene precipitation concerning temperature, the study will help in the prediction of the expected trend of asphaltene precipitation for different types of crude oil under field conditions when the temperature will change during production.