Assisted history matching using artificial neural network based global optimization method – Applications to Brugge field and a fractured Iranian reservoir Toomaj Foroud a , Abbas Seifi b,n , Babak AminShahidi c a Department of Petroleum Engineering, Amirkabir University of Technology, Tehran, Iran b Department of Industrial Engineering, Amirkabir University of Technology, Tehran, Iran c Institute of Petroleum Engineering, College of Engineering, University of Tehran, Tehran, Iran article info Article history: Received 4 January 2014 Accepted 24 July 2014 Available online 21 August 2014 Keywords: artificial neural network history matching global optimization brugge field abstract Reservoir simulation is a powerful predictive tool used in reservoir management. Constructing a simulation model involves subsurface uncertainties which can greatly affect prediction results. Quantifying such uncertainties for a field under development necessitates history matching that is a difficult inverse problem with non-unique solutions. History matching is used to minimize the difference between the observed field data and the simulation results and requires numerous simulation runs. In many engineering simulation-based optimization problems, the number of function evaluations is a prohibitive factor limited by time or cost. History matching in hydrocarbon reservoir simulation is one of such computationally expensive problems which pose challenges in the field of global optimization. One way to overcome this difficulty is to use an artificial neural network (ANN) as a surrogate model. This article presents an ANN-based global optimization method that is used for history matching problem. The method has been applied to an Iranian fractured oil reservoir and the famous Brugge field benchmark. Computational results confirm the success of this method in history matching. We compare history matching results obtained by the proposed method with those of manual history matching and those obtained by simulation based direct optimization algorithm. The results compares favourably with manual history matching in terms of matching quality. The proposed method is superior than the simulation based direct optimization algorithm in finding multiple matched scenarios in less computation time. & 2014 Elsevier B.V. All rights reserved. 1. Introduction Reservoir simulation is a powerful predictive tool used in reservoir management. One of the difficulties in reservoir simula- tion is the absence of reliable data on reservoir characteristics. The reliability of production forecasts obtained by reservoir simulations strongly depend on the proper calibration of the reservoir simulation model. It is generally accepted that any model used for predicting unknown future quantities should be able to reproduce known history data. History Matching is a part of model validation process and is a cumbersome and time consuming task due to the needs for numerous simulation runs. History matching workflow focuses on calibrating a reservoir model using observed dynamic data (e.g. production data) as well as measured static data (e.g. core or well log data). It requires solving an inverse problem for which the solution would be non- unique since many combinations of parameter settings would yield a similar model response. From an optimization perspective, the history matching problem can be stated as follows: min ‖OðxÞ y‖ 2 x A Ω ð1Þ where y A R m is the vector of measured observations and OðxÞ A R m denotes the simulated results. Here x represents reser- voir uncertain parameters which belongs to feasible domain Ω. The objective of this inverse problem is to find an x such that the distance between the resulting simulation outputs and the observed data is minimized. It requires excessive costly simulation runs, and in many cases the derivative information is expensive to obtain or may not be available. Various reservoir parameters such Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/petrol Journal of Petroleum Science and Engineering http://dx.doi.org/10.1016/j.petrol.2014.07.034 0920-4105/& 2014 Elsevier B.V. All rights reserved. n Correspondence to: Amirkabir University of Technology, 424 Hafez Avenue, P.O. Box 15875-4413, Tehran, Iran. Tel.: þ98 21 64545377; fax: þ98 21 66954569. E-mail addresses: toomaj_foroud@aut.ac.ir (T. Foroud), aseifi@aut.ac.ir (A. Seifi), aminshahidy@yahoo.com (B. AminShahidi). Journal of Petroleum Science and Engineering 123 (2014) 46–61