Machine Learning-Based Prediction of Surface Equipment Requirements for Well Tie-In Operations

Abstract

Well tie-in is a critical process for connecting new wells to the production system, maintaining flow assurance and system stability. The increasing complexity of reservoirs, surface facilities, and real-time operational constraints has made tie-in optimal planning more challenging. One primary step is to plan all surface equipment required for the tie-in, as missing any equipment will cascade into delays. This paper presents a novel machine learning (ML) engine that predicts all required surface equipment prior drilling. The paper discusses a pipeline of two ML models to predict the required tie-in surface equipment. The first model clusters existing wells into groups based on their geographical location and field characteristics using a technique called DBSCAN, which is a type of unsupervised learning that groups similar data points together. The second model is then trained using a classifier chain technique along with a classifier algorithm on these clustered wells to produce a multi-label prediction for the required equipment. To further improve the clustering accuracy, additional data dimensions were generated using Haversine distance technique and Cartesian map projection coordinates. The developed ML models were successfully integrated and used to predict the required surface equipment for well tie-in operations, including flowmeters, H2S detection systems, corrosion inhibition systems, and other essential components. Among all tested algorithms, Random Forest model delivered the highest performance, achieving an overall prediction accuracy of approximately 90%. The model consistently captured equipment patterns across diverse scenarios. Operational benefits were clearly observed. Automated early predictions helped avoid costly rework and unnecessary storage fees. Additionally, the automated prediction workflow reduced labor hours, minimizing manual analysis time and accelerating the tie-in preparation process. This paper presents a novel approach in expediting the well tie-in process by utilizing the pattern of clustering geospatial data in a model. The approach resulted in enhancing and expediting the well tie-in planning by predicting the appropriate surface equipment and reducing human input errors, which resulted in saving time and resources.