Uncertainty Assessment in Production Forecasting and Optimization for a Giant Multi-Layered Sandstone Reservoir Using Optimized Artificial Neural Network Technology

Abstract

This paper presents an uncertainty assessment project using Artificial Neural Network (ANN) for a giant multi-layered sandstone reservoir in Middle East, which contains several uncertainties and associated risks. Uncertainty quantification in history matching, production forecasting and optimization approaches often requires hundreds of thousands of forward flow simulations to explore the uncertain parameter space, causing forbidden computational time requirement, especially for large-scale reservoir models. In order to bypass this limitation, one can use a proxy to replace the time-consuming flow simulator. In this work, an optimized ANN is used as the proxy and an uncertainty assessment workflow is implemented for the giant Cretaceous multi-layered sandstone reservoir using a global optimizer. Using the ANN based uncertainty assessment framework, the impacts of the main uncertain parameters on production forecasting are assessed for this multi-layered sandstone reservoir. Then, field development optimization is also performed to optimize wells injection and production rates to maximize the economic measures considering uncertainties. Using a real field as a case study, the capability of our uncertainty assessment workflow is demonstrated. The optimized ANN successfully captures the complex nonlinear dynamics as a proxy and significantly reduced the computation time.

Oil Production Forecasting with Uncertainty Description Using Data Driven Proxy Model

“…Input datasets play a great role in studying uncertainty in production prediction. In addition, the uncertainty of input parameters makes it necessary to use a probabilistic approach [4], [6]. Although there are different methods for probabilistic data sampling, Latin Hypercubic Sampling (LHS) was chosen as an efficient sampling method [14].…”

“…These analytical models have been trained and developed to behave like simulators while consuming less time. In this way prediction, analysis, and finally optimization will be performed more efficiently [6], [7]. So far it can be mentioned that accuracy and acceleration are two important characteristics that should be considered during the reservoir simulations.…”

Accelerating Physics-Based Simulations Using End-to-End Neural Network Proxies: An Application in Oil Reservoir Modeling

“…It seems important to incorporate features of the reservoir. Examples of basic, fully connected ANNs to predict reservoir production are Wei et al ( 2017 ) and Wang et al ( 2018 ). In the context of pre-existing literature, we believe our contribution is twofold: (1) our architectural approach is unique as it deals with sequential action input (i.e., varying timing and locations of wells) and output of varying span, reservoir uncertainty, and optimized well control, while modeling the simulator in an end-to-end manner, and (2) we present a thorough experimental analysis on a publicly available reservoir model thus creating a reference for future comparison by the community.…”