Unravelling Favorable Mesozoic Reservoir Facies by Advanced Logging and Machine Learning, Offshore Bohai Bay China

Abstract Mesozoic oil and gas reservoirs have become a hot spot for exploration and evaluation in recent years as big potential reserve replacement in offshore Bohai Bay, Eastern China. The complex tectonic movements, multiple uplifts, and various lithologies (volcanic, clastic and igneous) result in different degrees of reservoir diagenesis and strong heterogeneity in fracture development. Rapid identification of favorable lithofacies and effective natural fractures is crucial for reservoir evaluation in time-cost saving manner. Unlike conventional clastic reservoir, the Mesozoic reservoirs in the Q field show low matrix porosity and permeability. The field mud gas data and cuttings were affected due to multiple drilling stuck and mud loss, which can't provide continuous and accurate hydrocarbon indicator. Advanced logging suites including dual borehole image logs and Nuclear magnetic resonance (NMR) logs were deployed to achieve this. Porosity and permeability were firstly analyzed based on core analysis and formation sampling from previous appraisal wells, showing that porosity ranges from 2.5% to 15.3% and permeability ranges from 0.085mD to 34.5mD. Traditional permeability equation was based on the regression between core porosity and permeability. However, for lithologies of different grain sizes, the porosity-permeability relationship is very confusing, and a single regression cannot reveal the correct relationship between them. Classifying lithology by grain sizes and minerals and establish the porosity and permeability relationship is also not working properly due to the very low permeability. Therefore, the Stoneley wave property is applied to estimate the permeability of these reservoirs, which is proved to be more applicable. NMR porosity and permeability were calibrated by core data to get continuous analysis. Both electrical image and ultrasonic image were used to identify fractures and calculate fracture parameters. Based on the amplitude image and travel time image, five types of fractures are identified: closed fractures filled with clay, closed fractures cemented by calcite/quartz, fully open fractures, partially open fractures, and vuggy open fractures. A 2D indexed and probabilized self-organizing map (IPSOM) was applied for favorable reservoir facies clustering. Totally eight sensitive variables were selected, NMR permeability, effective porosity, and secondary porosity, compressional slowness, fracture aperture, and fracture length, fracture porosity and fracture density. And eight reservoir facies were classified from IP1 to IP8. IP1 is tight facies, IP2 is high porous – strongly vuggy facies, IP3 is medium porous – fractured facies, and IP4 is medium porous – fractured facies, IP5 is fractured – tight facies, and IP6 is high porous – fractured facies, and IP7 is fractured strongly vuggy facies, and IP8 is vuggy – high porous facies. The reservoir quality of IPSOM facies from the best to the worst are: IP6>IP8>IP2>IP7>IP4>IP3>IP5>IP1. The results were used as supervised machine learning inputs to train other wells, and favorable reservoir facies were analyzed more efficient than before. The advanced logging suites and machine learning based reservoir facies provides a reliable and solid reference for subsequent development well deployment and unravel the potential reserves in the Mesozoic fractured reservoir. It is crucial to field development strategy and production enhancement.