CAUSAL MACHINE LEARNING FOR IDENTIFYING ROOT CAUSES OF PRODUCTION DECLINE: DISTINGUISHING OPERATIONAL AND RESERVOIR DRIVERS

Analysis of the decrease in production is a severe part of the petroleum asset management as the correct determination of the causes which are behind the decrease is a crucial thing to optimize the performance of the field and decrease the useless interventions.The conventional approaches, like empirical decline curve analysis and physics-based reservoir simulation, may not always capture the interactions between the operational parameters and the reservoir behavior adequately, and hence they cannot distinguish between the operational and subsurface causes of the reduction in production.ThIS paper uses Causal ML methods, causal forests and causal treatment-effect estimation, to characterize the decline in production in a given oil field and evaluate the importance of various causal factors.The parameters that are evaluated on an operation basis are choke settings, frequency of the downtime events, artificial lift efficiency and change in the surface facility pressure.Specific indicators that were studied in reservoirs consist of water cut development, rate of depletion of reservoir pressure, degradation of permeability and coning.With the counterfactual analysis, the research removes the influence of each variable on the measured production patterns.These findings reveal that operational bottlenecks can be a contributor to around 62% of the decline in production, and 38 percent is triggered by a mechanism related to the reservoir.These results indicate that Causal ML can offer a more reliable and practical diagnosis framework than its conventional counterparts, as it can give engineers an idea of the possibility of in-service operational conditions and implement specific measures.In general, the paper indicates that Causal ML can help to enhance the quality of production surveillance, the effectiveness of operations of the currently existing digital oil field environments, and contribute to making decisions that are supported by data.