| The development of unconventional reservoirs relies extensively on Massively Hydraulically Fractured Horizontal Wells (MHFHW) to produce economic production rates. However, as the desired half-length of the massive hydraulic fracture increases, its effective half-length and level of confinement decrease while its associated cost increases. These aforementioned disadvantages become more pronounced in ultra-tight and naturally fractured formations, such as shale gas formations. On the other hand, Complex Wells (CW) are becoming an attractive alternative to MHFHW in unconventional formations. CW have an extended reach and advanced structures that increase their contact with the reservoir, thus enhancing their productivity. The field application of CW in unconventional formations is increasing, especially with technological advances in directional drilling and geo-steering. With such an increasing architecture complexity of CW and the heterogeneous nature of shale gas formations, a quick yet robust expert system is needed to predict their production profiles, analyze their transient data, and design their well configurations.;Developing an analytical model to predict complex CW performance is an arduous task, especially in unconventional formations. In addition, the mere use of numerical modeling to design a well that matches a desired production profile requires significant time investment to model all possible scenarios and then to optimize. Furthermore, traditional well testing methods do not capture the complex interaction between CW and shale formations. In addition, they require manpower, equipment, a lot of time, and capital investment. However, the Artificial Neural Networks (ANN) ability to recognize nonlinear relations that govern fluid flow in porous media and instantly predict results, makes it an attractive method. Therefore, in this study, several ANN are trained to perform three major tasks instantaneously. The first task is to predict CW performance of a given well configuration from a given shale gas reservoir. The second task is to predict the well configuration that will produce a desired production profile from a given shale gas reservoir. The third task is to predict the reservoir properties from a given production profile and well configuration.;A commercial numerical reservoir simulator is used to generate a database to train, validate, and test the ANN used in this study. For each ANN, the database becomes the source of input and output parameters, which include production profiles, well design parameters, well operating conditions, and reservoir properties. A new database is generated for each task. All input parameters are randomly generated within their natural maximum and minimum values to build the reservoir simulation cases. |
