A Neural Network Approach for Modeling Water Distribution System

Abstract The successful waterflooding field development depends not only of the understanding of reservoir characterization, sub-surface injection displacement and sweep-efficiency, but also an accurate and effective design and operation of the surface network’s water injection distribution. In certain cases, the latter is critical for the successful waterflood operations in large fields with thousands of wells and high volume of new development activity. For the purpose of this study, we are presenting a new data-driven approach to accurately estimate the injection rate in all non-instrumented wells in a large waterflooding operation. A collection of data driven tools, including statistics, clustering, simulation and an artificial neural network model were employed to prime and model the data. As a final point, the neural network leverages instrumented wells’ data and serves as an accurate real-time proxy to estimate missing injection rate measurements in non-instrumented wells. The system’s accuracy was validated by comparing the estimated rates for different wells on a different branch with the ones measured at physical wells. The neural network model trained on the cleansed data set revealed a high performance system with a 0.93 R2 values for both training and validation sets. The paper outlines both the methodology and procedures used to analyze a branch of the water network system, and the modeling of accurate estimation of injection rates. The model performance is remarkable having used only field and wellhead measured data and considering the natural uncertainty inherited in these values. Finally, this system provides the capability to estimate the flow rate for every non-instrumented well in the field and respond to exceptions in relevant time.