Unconventional reservoir prediction from well-logging data with time-series machine learning models: A study case from the second sub-member of the third member of the middle Triassic Leikoupo Formation, central Sichuan basin, SW China

The middle Triassic Leikoupo Formation in central Sichuan basin, China contains important dolomite and gypsum salt rock reservoirs that are prospective for oil and gas exploration. However, a recent discovery of a prolific oil flow in the marl limestone and limy mudstone of the second sub-member of the third member of the middle Triassic Leikoupo Formation. Influenced by intermittent sea erosion and confined evaporation, the second sub-member of the third member of the middle Triassic Leikoupo Formation in the middle of Sichuan basin lagoon was mainly deposited as an interbedded layer of plutonic saltstone and marl limestone, depositing many sets of black organic-rich marl limestone with strong hydrocarbon production capacity, indicating additional unconventional reservoir potential. A study was conducted on the marl limestone and limy mudstone reservoirs in this sub-member. An integrated lithology - electrical - depositional sequence response model was developed based on a long short term memory (LSTM) model. The model was trained on logging and lithology data from 27 wells, then used to predict lithology for 24 additional wells. It predicted marl reservoirs in 13 specific wells, indicating potential for further hydrocarbon discoveries. Comparison with core and petrographic data could help quantify the prediction accuracy and determine key geological controls on reservoir distribution. Ultimately, based on the predicted results, the centre of unconventional reservoir development in he second sub-member of the third member of the middle Triassic Leikoupo Formation in central Sichuan basin was clearly defined and the lake basin developed in a narrow band. The data-driving model enabled more accurate prediction of unconventional reservoirs in this sub-member and achieved fine grained exploration guidance.