Case Study of Applied Artificial Neural Networks on Forecasting the Essential Performance in a CO_2 Enhanced Oil Recovery Process

摘要

Among various enhanced oil recovery (EOR) technologies, CO_2 flooding is utilized as an effective method because it can remarkably improve oil production in depleted oil fields and is environmentally associated with a geological carbon capture and storage (CCS) project for reducing carbon emissions from anthropogenic sources. The use of hybrid smart tools to predict the uncertainties of CO_2-EOR projects has been suggested by previous research, but their use is still restricted on forecasting the overall performance during flooding processes. The objective of this work was to generate and investigate the applicability of network models to forecast the essential performance of water-alternating-gas (WAG) injection in terms of the oil recovery factor (RF), oil rate, gas oil ratio (GOR) and net CO_2 storage for a five-spot pattern scale. In total, 239 numerical samples were simulated and collected to train the networks. The abovementioned outputs were predicted after 10, 20, 30 and 40 cycles of injection in each network following changes of the initial water saturation, vertical-to-horizontal permeability ratio, WAG ratio and duration of each cycle. The generated network models for the oil recovery factor and CO_2 storage had excellent accuracy in estimations with an overall root mean square error for both models of less than 3%, while the errors for the oil rate and cumulative GOR fluctuated from approximately 5% to 7%. Because the deviation of the GOR and oil rate among cases was significantly high, it is recommended to predict the GOR after 20 injection cycles and the oil rate after 30 cycles. The relative correlation representing the physical terms between the design variables and outputs were also determined. The large difference in the dimensions of permeability was favorable for both CO_2 storage and final oil recovery; a WAG ratio lower than two was recommended for the flooding design. Unstable reaction relationships among parameters during the injection were observed, indicating a complication of evaluating a CO_2 flooding process in enhanced oil recovery.