Sensitivity Analyses of Production and Recovery Forecasts Using Variance-Based Methods

Abstract

Abstract During the exploratory stage of field development, reservoir data required for forecasting is often inadequate, and hence uncertain. An understanding of the scope and extent of uncertainty is invaluable in optimizing data gathering operations, planning of field facilities, and assessing exploration and development scenarios. Sensitivity Analysis (SA), the study of how the uncertainty in a model's outputs depends on the uncertainty in its inputs, provides such an understanding. This work presents the SA of production and recovery histories using a new continuous time Monte Carlo (MC) multiwell tank model (UTSens). The model assumes that reservoir parameters are independent; knowledge about a parameter, does not affect knowledge about other parameters. The model also assumes that the reservoir is homogeneous and undersaturated, and that the reservoir is produced by primary depletion, specifically volumetric expansion. The MC approach is probabilistic rather than deterministic; as such it computes the mean, variance and sensitivity histories of reservoir and well production rates, recovery, and pressure. The model provides a choice between Latin Hypercube and Sobol LPT sampling. These sampling methods are more efficient than conventional MC or random sampling. In the cases considered, the sensitivity effects of different parameters vary over the producing period. Parameters with negligible main effects nevertheless can have significant joint effects. In one of the cases considered, the main effects of oil viscosity on production rate declined over twenty years of production with a single well. Conversely, the main effect of reservoir thickness increased during the same period. Other parameters like permeability and initial reservoir pressure with no main effects showed appreciable joint effects. In its totality, the current work is part of a paradigm shift from deterministic estimates to considerations of probabilistic analysis and uncertainty. This prescription is especially relevant during reservoir exploration and appraisal. Data analysis using the MC tank model provides metrics of uncertainty for optimized data gathering and informed decision-making.