AI Innovations in Waterflood Management, The Path to Autonomous Operations

Abstract

Abstract In the S-Field operations office, a daily battle ensues in the quest to increase production and maximize profits from waterflooding. One of the main control mechanisms applied to optimize the waterflooded reservoirs is by controlling the water injection and pumping rates of producers to balance patterns, maximize sweep, and maintain reservoir pressure. The reservoir surveillance team has been using a simple spreadsheet analytical approach that was quite limiting as the number of injection patterns increased, and the flood matured, leading to a water breakthrough. There was a need for a more sophisticated approach that could leverage artificial intelligence (AI) technology, especially since the entire asset was undergoing significant digitalization of its operations. This paper presents various innovations in bringing real applications of AI for waterflood management. This includes innovations in business processes, application of design thinking methodology, agile development, and AI. The AI waterflood management solution combines cloud technologies, big data processing, data analytics, machine learning algorithms, robotics, sensors and monitoring system, automation, edge gateways, and augmented and virtual reality (AR/VR). Design thinking principles and a human-centric approach within an agile innovation framework were utilized for rapid prototyping and deployment. A waterflood management framework that addressed the business's operational, tactical, and strategic aspects created the backdrop for designing the solution architecture. New injector-producer modeling techniques that leveraged AI and were fit-for-purpose for reservoir surveillance and production engineers were prototyped. An interactive pattern flood management tool, adapted from streamline simulation-based waterflood analysis methods, was developed for injection pattern analysis and intelligent optimization workflow. Field pilot testing for over a year proved that the prototype could reliably detect injector-production interactions and recommend operating set points in relevant time. Reduced time to decision, improved analysis efficiency and reliability of short-term forecasts, reduced field visits and health-safety-environment (HSE) exposure, and finally ease-of-use has been experienced. The learnings from this project are being leveraged to develop a deployable solution and move the needle toward autonomous waterflood operations.