Developing A Grid-Based Surrogate Reservoir Model Using Artificial Intelligence Shohreh Amini Reservoir simulation models are the major tools for studying fluid flow behavior in hydrocarbon reservoirs. They are now being used extensively in performing any kind of studies related to fluid production/injection in hydrocarbon bearing formations. Reservoir simulation models are constructed based on geological models, which are developed by integrating data from geology, geophysics, and petro-physics. This data comes from observation, measurements, and interpretations. Integration of maximum data from geology, geophysics, and petro-physics, contributes to building geologically complex and more realistic models. As the complexity of a reservoir simulation model increases, so does the computation time. Therefore, to perform any comprehensive study which involves thousands of simulation runs (such as uncertainty analysis), a massive amount of time is needed to complete all the required simulation runs. On many occasions, the sheer number of required simulation runs, makes the accomplishment of a project's objectives impractical. In order to address this problem, several efforts have been made to develop proxy models which can be used as a substitute for complex reservoir simulation models. These proxy models aim to reproduce the outputs of the reservoir models in a very short amount of time. In this study, a Grid-Based Surrogate Reservoir Model (SRM) is developed to be used as a proxy model for a complex reservoir simulation model. SRM is a customized model based on Artificial Intelligent (AI) and Data Mining (DM) techniques and consists of several neural networks, which are trained, calibrated, and validated before being used online. In this research, a numerical reservoir simulation model is developed and history matched for a CO2 sequestration project, which was performed in Otway basin, Australia where CO2 is injected into a depleted gas reservoir through one injection well. In order to develop SRM, a handful of appropriate simulation scenarios for different operational constraints and/or geological realizations are designed and run. A comprehensive spatio-temporal data set is generated by integrating data from the conducted simulation runs and it is used to train, calibrate, and verify several neural networks which are further combined to make the surrogate model. This model is able to generate pressure, saturation, and CO2 mole fraction at each grid block of the reservoir with a significantly less computational effort compared to the numerical reservoir simulation model.