Development of an Artificial Neural Network for Cyclic Steam Stimulation Method in Naturally Fractured Reservoirs

Open Access
Arpaci, Buket
Graduate Program:
Petroleum and Natural Gas Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
April 17, 2014
Committee Members:
  • Turgay Ertekin, Thesis Advisor
  • Artificial neural network
  • cyclic steam injection
  • heavy oil
  • naturally fractured reservoirs
In recent years, increased demand of energy requires improvement in recovery factor by implementing enhanced oil recovery (EOR) method that can be applied efficiently into the naturally fractured reservoirs. Cyclic steam stimulation (CSS) is one of the thermal EOR techniques in which reduction of oil viscosity is achieved by increasing the reservoir temperature. A single well is applied in the CSS process. The well is used for both injection and production, sequentially. The implementation of CSS into the naturally fractured reservoirs can be considered more appropriate because fractures provide a large area in where injected steam diffused. In the petroleum industry, commercial simulators are applied for forecasting of production profile; however, their algorithms are considerably complex and time-consuming. Usage of artificial neural network (ANN) is getting attention in recent years owing to its ability to provide a solution for non-linear relationship. In this study, an elliptical inner zone that is more fractured than corresponding naturally fractured reservoir is designed to increase the performance of CSS process. A variable number of cycles is studied and production period of each cycle is controlled by specified abandonment oil flow rate. The goal of this study is to provide an accurate estimation for the performance of cyclic steam injection process in relatively short period by developing artificial neural network models. Design parameters of both fractured inner zone and cyclic steam injection are used as variables of ANN models, besides reservoir properties. Production profiles of the first ten cycles of each case are evaluated during the network training. Six ANN models are developed and a total of 555 case samples are generated in order to train the networks. Two of them are assigned to forward-looking problem. Forward ANN-1 is designed as a predictor of oil flow rate and number of cycle whereas Forward-ANN-2 is constructed as a predictor of cumulative oil production of project, cumulative oil production and cycle duration of each cycle. Two of the six ANN models are generated for estimation of design parameters by using performance indicators and properties of corresponding reservoir. Inverse ANN-1A is able to predict fractured inner zone and cyclic steam injection design parameters at the same time. Inverse ANN-1B is developed to improve the capability of design parameters prediction by focusing on only cyclic steam injection design. The last two networks are assigned to prediction of reservoir properties by applying desired production profile and design parameters. Inverse ANN-2A is created for estimation of reservoir properties after fractured inner zone was created, in case performance indicators and CSI design parameters are provided. Inverse ANN-2B is trained as a predictor of only fractured inner zone properties. Significant improvement is observed in terms of accuracy by developing Inverse ANN-1B and Inverse ANN-2B additionally.