Artificial Intelligence and Machine Learning Techniques Provide Operations Geologists With an Automated and Reliable Lithology-Fluid Pattern Recognition Assistant: A Case History in a Clastic Reservoir in West Africa

AI is beginning to take on a key role in assisting geologists during well data collection and analysis. Thanks to AI assistants, geological well data can be quality controlled, cleaned, prepared and then paired with preliminary pattern tags in order to automate interpretation processes. Geologists still remain in charge of reviewing and consolidating AI provided interpretations, instead of interpreting raw data from scratch. This paper describes how AI supported the interpretation of lithology and fluid type in a Cretaceous clastic reservoir in West Africa. An AI assistant supported by AI algorithms has been trained to recognize five different lithology-fluid patterns of interest: Claystone, Sandstone-Oil, Sandstone-Gas, Sandstone-Water and Limestone-Water. Training and validation phases have been run against data from more than 20 different wells. Data gathered consisted of few tens of input variables, spanning from LWD logs, to GAS variables and drilling parameters. Overall, the training dataset consisted of more than 125k examples, corresponding to approximately 19 kms of geological sampling. Several algorithms and modelling strategies have been evaluated within a recursive prototyping framework. Gradient Boosting, Random Forest and Neural Networks supervised machine learning algorithms, coupled with both unsupervised and optimized hyper - parameters tuning approaches, have been compared and selected to achieve better tagging performances. Gradient Boosting algorithm, when supported by hyper-parameter tuning based on genetic algorithm, results to outperform Random Forest and Neural Networks, together with any alternative approach among those evaluated. More specifically, tagging accuracy achieved on both, validation and holdout samples, demonstrates an encouraging understanding of Sandstone-Oil, Sandstone-Gas, Sandstone-Water and Claystone. This alternative AI enabled workflow – when applied in real time - could provide a significant added value by decreasing the time needed to identify the best strategy case-by-case and consequently the time-to- production for each new well.