No abstract

In this paper, a new approach was identified and tested to detect abnormal events in producing wells when a labeled dataset is unavailable or the number of instances are below 10% and are insufficient for conventional modelling methods. Autoencoders (AE), a type of unsupervised learning, are trained to learn normal behavior by trying to reconstruct the input data that is fed into the model. When run in prediction mode, low reconstruction errors are classified as Normal behavior whilst higher errors are classified as anomalous behavior. Different model structures were tested. An average accuracy of 94% with a precision and recall rate of 70% was achieved using a 6-Layered AE-NN model. The results of the models created show encouraging results and can help detect events and notify engineers when the well is deviates from expected behavior.

Efficient reservoir management is defined by its workflow structure stretching from a comprehensive integrated reservoir characterization (static and dynamic) to fast and reliable decision making for optimal field development planning. A lengthy and fragmented process that is scattered among multiple disciplines and data domains. The oil industry has made tremendous effort in advancing technologies to enable faster and smarter subsurface modeling, history matching, scenario prediction and risk analyses. Achieving optimality with reduced-risk requires typically prohibited high amount of human effort and technology resources. As we enter into the Fourth Industrial Revolution of Big Data, Artificial Intelligence (AI) and Internet of Things (IOT) devices, alternatives complimentary tools to the conventional techniques have surfaced. The purpose of this work is to demonstrate the advantages and shortfalls of commercially available data-driven model techniques for dynamic reservoir forecast as compared to traditional numeric simulation techniques. In this paper, a consistent procedure to compare the performance of both data-driven models and full-physics numerical models is developed and discussed. The procedure was tested on an onshore reservoir located in Abu Dhabi, UAE that provided the basis and results for the analysis. The approach included analysis of human resources requirements, computer hardware, simulation performance and models updating strategy.

Asset management success is accomplished when the integrated production system is operating close to its intended potential. Continuous awareness of wells and facility conditions are key factor in the realization of designed capacity. In contrast, unknown status and conditions can severely limit production capacity. The rise of instrumentation technologies over the last four decades have created new opportunities to understand well and reservoir behavior. However, despite of being proved as a cost-effective surveillance initiative, remote monitoring is still not adopted in over 60% of oil and gas fields around the world. Understanding the value of data through machine learning techniques is the basis for establishing a robust surveillance strategy. The objective of this paper is to develop a data-driven approach, enabled by Artificial Intelligence (AI) methodologies including machine learning (ML), to find optimal operating envelope for gas-lift wells. The process involves building ML models for generating instantaneous predictions of multiphase flow rates and other quantities of interest, such as GOR, WCT, using real-time sensor data at the surface, historical performance, and sporadic test data. Additionally, forecasting models were developed for generating short-term (30 days) forecast of cumulative oil, water, gas, and liquid production, multiphase flow rates, WCT, GOR, and reservoir pressure. Using time-series forecasting models, a sensitivity analysis was performed to generate short-term well response for a selected number of combinations of choke settings, and gas injection rates. Sensitivity analysis provides 2D maps of well response highlight an operating envelope, which are proposed to be combined with physical and operational constraints to arrive at optimal operating conditions, which may effortlessly add 2.5% net profit from optimum gas-lift alocation. The results of this work show encouraging results, and demonstrate value that AI-enabled methodologies can provide in instrumented wells by enabling automated workflows for virtual metering, production allocation, short-term production forecasting, and deriving optimal operating conditions. The developed AI methodology has tremendous potential of integration in an end-to-end workflow of autonomous well control by utilizing available data to produce easy to update ML models, with little to no human intervention.