Development of artificial expert reservoir characterization tools for unconventional reservoirs

With the decline in production from conventional hydrocarbon resources, new focus has been shifted to unconventional resources. However, oil and gas production from these types of hydrocarbon resources is not as easy as producing from the conventional resources because of the complex geological features and lack of new technologies. Soft computing techniques such as artificial neural networks provide new approach as that can be used in the characterization of the complex unconventional reservoirs. In this study, artificial expert systems were developed with the purpose of characterizing an unconventional oil reservoir located in West Texas. These expert systems are capable of generating synthetic well logs, completion parameters, production profiles and performing the task of payzone identification. This study focuses on the generation of synthetic well logs and the identification of payzones using artificial expert systems. Synthetic well log prediction module is divided into low-resolution and high-resolution categories where five different well logs are predicted at desired reservoir locations. While low-resolution well logs are predicted using the averaged seismic data, the high-resolution well logs are predicted using detailed 3D seismic data. Training of the networks to predict high-resolution well logs is found to be more successful than that of low-resolution well logs. Predicted synthetic well logs are then used to predict completion data, production profiles and payzone identification.;The second module of this research involves payzone identification in which the gross thickness of the reservoir is ranked based on its productivity level. Payzone identification is achieved through the implementation of artificial expert systems developed to predict well performance (i.e. oil, water, and gas production profiles). Using a moving-window approach to sample seismic and well log data along the well depth and by feeding the sampled information to the well performance network, it is possible to predict the productivity of each sampled segment. Another outcome of the payzone identification study is the possibility of scrutinizing the relationship between well log parameters and expected productions. A Fuzzy classification method is used to classify production data in terms of lithology logs. One of the outcomes of this classification is the realization that oil production is expected to be higher in shaly segments of the well than that of carbonate segments.