Search

Pore pressure (PP) prediction from the downhole response is challenging due to the complex relationship between PP and the underlying variability of downhole response at continental slope setting. Conventional methods are mostly deterministic, and they do not usually take into account the underlying variability and uncertainty in the prediction. Here, we implement statistical machine learning (ML) augmented interpretation techniques for the prediction of PP of well 1344A located at the slope setting of sites IODP 323. Within this framework, we quantify the relative influence of each downhole response and hydrostatic pressure (HP) on PP prediction by taking, (i) downhole responses as input and PP as a target, and (ii) downhole responses + HP as input and PP as target and to fix input parameter lag by examining partial auto-correlation function (PACF) analysis results for ML model development. Comparative performance evaluation shows Gaussian process-based GPRN model is superior to the other three ML models such as ANFIS, ARD-BNN, and SVM based on R2 value between the target via Eaton's and porosity methods and predicted via ML models. Uncertainty analysis shows that GPRN-based prediction falls within the 95% confidence interval (CI) at most of the depths. Further, the derivation of ML-based polynomial equations can be an interesting step towards the development of data-driven prediction of PP that potentially would be exploited for data-driven PP prediction studies in other complex geological settings of the world. [ABSTRACT FROM AUTHOR]