Complex Network Method for Inferring Well Interconnectivity in Hydrocarbon Reservoirs

This study presents an ad hoc machine-learning approach to uncover dynamical correlations among production and injection wells in naturally fractured reservoirs. The method is an application of the Visibility Graph Analysis (VGA) technique introduced by Lacassa et al. (2008) to analyze the state of an oil field. Due to its statistical nature, this technique relies on the available field information and historical production logs, makes no assumptions, and does not require specific information about the reservoir’s properties. We assume that all inter-well’s dynamic details are fully encoded in the historical logs. Thus, all production-injection data is transformed into a set of complex multiplex networks that take advantage of the VGA technique at the method’s core. These sets of networks map the production rates onto a ‘state’ image of the well system at a particular time. One well’s influence on another is then established by assessing the time development of the respective states, thereby effectively capturing the long and short-term dependencies and revealing the evolution of the well’s ‘hidden’ interconnections. This methodology was applied to a dataset (whose logs span sixty years) belonging to an active oilfield in Mexico. Our results show that the transitions between the field’s states provide crucial insights into the flow pathways within the reservoir. Moreover, they give a clear picture of how the operation of certain wells influences the production of other wells (e.g., when the production of oil in a well increases, decreases, or when a well switches from a production to an injection operation).