Automated Hydraulic Fracturing Integrated with Predictive Machine Learning

Abstract

Abstract In unconventional wells, returns are driven in part by the reduction of variability in efficiency and performance. In 2021 the stimulation of two wells in the Bakken proceeded under the architecture of an automated frac control system communicating directly to Machine Learning predictive models at the headquarters of the well operator. This represents the first time an algorithmic frac was conducted via automation, adjusting stage designs, and pushing those stage designs to the field without human intervention. This paper discusses a fully integrated and automated completion performed on the operator's four- well pad in the Bakken. It reviews the impact on completion performance, completion design, components of the system and execution. Throughout the completion, automated software interfaced with the frac control system executing the job. Additionally, data was uploaded live and fed to the Machine Learning predictive model. This allowed the model to learn from actual well data and suggest improvements. Improvements were captured, iterated on, and design updates were sent back to the control system for the next stage in the completion sequence. Human oversight was conducted but only as a check, during the entire process. Both the automated frac control system and algorithmic design system were functionally separate but communicated live, allowing the operator to take advantage of their complete basin knowledge database without compromising data integrity and model confidentiality. Additionally, sensors provided real-time data such as treating pressure, rate and proppant concentration, as well as downhole data such as cluster uniformity, fracture geometry, and offset well interactions. The project was launched with several primary goals in mind: First was to functionally test the automation of the frac fleet for the operator proving its ability to consistently place their designs. Second was to incorporate the prediction model algorithms into completion design and test how quickly and how much the Machine Learning models could actually learn from actual well stages. Both of these primary goals were achieved, validating the ability to automatically execute completions and to tie design changes live to a control system elsewhere. This represents the first time a hydraulic fracture was conducted via automation with algorithmic integrated design improvement, either independently or together. These capabilities can improve execution and performance where it is becoming increasingly difficult to deliver step changes in well performance with current manual crews and technology. Integrated automation provides an upgrade to completion performance by reducing variability in execution and well performance while also enabling tailored designs on scales previously unattainable.