Continuous Steam Distribution Optimization using Pattern Recognition on High Frequency Data Streams

Abstract

In this paper, an assessment of a streaming dataset from all active steam injectors in a mature steamflood field is carried out to understand and identify data trends and patterns which indicate if a steam injector is out-of-design. The dataset utilized in this study comprises real-time data and data in motion available thanks to the newly instrumented asset. However, this high-frequency data, while available, was never analyzed before this study. This work showcases the first study of this kind that utilized high-frequency streaming data from steam injectors. As an exploratory study, it revealed powerful insights and patterns which explained not yet understood behaviors. The methodology employed involved management and analysis of large volumes of data and consideration of the steam distribution system network. The study revealed the root causes of out-of-design and questionable steam quality values, which led to a comprehensive report describing the events, recommendations, and remedial actions for 33 out of 111 active injectors. The business driver for this project relies on solving cases in which the real steam quality is unknown, or the injector is out-of-design, which affects the steam-flood delivery and, consequently, the oil production performance. The novelty of this study relies on the capability of identifying undesired events at very early stages. In similar oilfields under steam-flood operations, steam injectors performance is tested and analyzed every three to six months. Many undesirable events may occur and are ignored in the time window between tests. The study not only led to business value impact due to addressing un-optimized injectors but also started a new program for real-time monitoring. This research demonstrates the value of using high-frequency raw data for steam injectors diagnosis, management, and monitoring.