Analysis of Gas-Cycling Performance in Gas/Condensate Reservoirs Using Neuro-Simulation

Abstract

Abstract The study of gas-condensate reservoirs has been a fertile area of research in the last decades, especially because of their singular depletion behavior which cast them apart from other types of oil and gas reservoirs. Typically, exploiting a gas-condensate reservoir without pressure maintenance generates high surface gas recoveries with relatively low condensate recoveries. Appearance of a hydrocarbon liquid phase upon isothermal depletion is typical of these systems, and major concerns arise related to the loss of this valuable fraction of heavier hydrocarbons and the associated impartment in gas production around the wellbore region. As a consequence, the injection of either surface gas or other gases such as nitrogen is common for pressure maintenance purposes. Pressure maintenance is economically justified by keeping the reservoir pressure above dew point or revaporizing any valuable condensate that might have formed. The convenience of the gas injection operation for a specific gas-condensate reservoir is strongly tied to the flow characteristics of the reservoir and phase behavior of the fluid. Highly sophisticated and expensive compositional simulations are usually employed in order to calculate reservoir performance under gas-cycling operations. In this study, we propose the implementation of artificial neural network (ANN) technology towards the establishment of an expert system capable of learning the existing vaguely understood relationships between the input parameters and output response of compositional simulation of gas-cycling operations in gas-condensate reservoirs. Parametric studies are conducted which identify the most influential reservoir and fluid characteristics in the establishment of optimum production protocols for the implementation of the pressure maintenance operation. As a result, a powerful tool for the implementation of pressure maintenance operations in gas-condensate reservoirs is developed. This tool is capable of assisting in designing an optimized exploitation scheme for a particular reservoir under consideration for development. Fast, reliable, and inexpensive results are obtained using the proposed approach while capturing the important flow and thermodynamic characteristics inherent to such systems without resorting to detailed and expensive compositional simulation studies. Introduction An artificial neural network consists of a number of highly interconnected neurons which provide a platform to perform highly non-linear, multidimensional interpolations between output and input variables. These interpolations can potentially reveal the relationship among these variables, regardless of the obscurity of the relationship among them, even for not well understood non-linear relationships between the input and output variables. The ANN technology was inspired by the workings of the human neurons, where much of the learning is accomplished by experience and repetition. The neuro-simulation technique targets the establishment of a powerful symbiosis between hard-computing (e.g., rigorous numerical computations) and soft-computing protocols (e.g., artificial neural networks). In neuro-simulation, hard-computing techniques are coupled with soft-computing techniques for the development of powerful expert systems. Numerical models provide a precise and formal "expertise"—at a significant computational expense—that can be taught to a soft-computing tool that, once trained, can exploit and apply the learned expertise through much less intense computational work.