چكيده به لاتين
In this thesis attempts to create and develop an intelligent data-driven model for screening and analyzing the EOR methods. Initially, data is gathered, processed and reviewed in two parts of the field and simulation section. The data based methods are discussed in general and based on the application. In the project data preparation, data processing, screening, forecasting parameters, production forecasting has been performed. Data preparation section includes normalizing, identifying outlies, summarizing data. A data-driven screening model (DDSM) is developed to screen the enhanced oil recovery (EOR) methods for petroleum reservoirs using combined capabilities of fuzzy expert approach (FEA) and support vector regression (SVR) techniques. In order to improve screening performance, a fuzzy model was integrated using 4 SVR models to predict the screening parameter’s effective weights. The SVR models can predict recovery factor (RF) of EOR methods including gas, chemical, steam, and combustion to calculate adaptive effective weight of the screening parameters. The SVR models were trained with datasets generated from simulations of EOR process. The absolute average error (AAE) of SVR models from simulation varied in the range of 0.078-0.095 for RF prediction. EOR field data of the past 40 years were reviewed to regenerate new and reliable EOR criteria table as basis for the screening model. The DDSM was evaluated to determine quantitative screening and ranking of EOR methods using seven field datasets, as well as fast forecasting of the nominated EOR methods. The DDSM results were compatible to published data in literature. In addition, the developed model can give comparable results with respect to EORgui software. The results show improvements due to adaptive weighting system on EOR screening methods for the studied reservoirs relative to fuzzy engine with constant weights. The presented model can guide the screening process to select efficient EOR method in practical applications.
The methods used in this research include evolutionary techniques, classifications, machine learning such as ANN, SVM, FL, SVD, PCA, Wavelet, GA, VQ,... which are combined or separately in different project phases on existing data. The results obtained in the primary, secondary and tertiary screening sections are compared with the actual data and the performance model is satisfactory compared to similar software such as EORGui and real data. The error in weight prediction model were between 3.7-13.5 percent and in reservoir simulation data based model were between 0.9-8.9 percent.
