An Unsupervised Learning Framework for Managing ESP-Driven Oil Field Operations with Water Injection in Northern Oman

This paper presents a framework utilizing unsupervised learning to optimize the management and productivity of an oil field in Northern Sultanate of Oman, operated by electric submersible pumps (ESP) and supported by water injection wells. By analyzing both historical and real-time data from 25 ESP wells and 27 water injection wells, the study evaluates operational efficiency and identifies underperforming wells. The approach aims to streamline field operations and drive performance optimization through data-driven insights. The data set of wells included continuous real-time monitoring of key parameters from ESPs and associated well infrastructure. These parameters encompass intake and discharge pressures, tubing head pressure, and flow line pressure. Additionally, the dataset integrates historical well test data, capturing vital metrics such as oil and water production rates, water cut, and gas-oil ratio. The proposed framework utilizes a two-stage analytical process. Initially, an exploratory data analysis was carried out for preliminary analysis, visualization and quality-check of collected data. Data imputation techniques were used to estimate missing data with different considerations for injectors and producers. Wells with abnormal behavior were identified by integrating anomaly detection, namely the isolation forests algorithm, into the exploratory analysis. In the second part of the study, k-means clustering was applied to categorize wells based on their operational efficiency and water injection performance. Elbow method, scatter plots and heat maps were used to evaluate cluster cohesion/separation quality. This methodological approach ensures precise and actionable categorization, enabling effective field management of ESP wells and water injectors. Applying k-means clustering to ESP and water injection well data separately segmented the wells into distinct performance-based clusters. This segmentation facilitates detailed operational behavior analysis and identifies specific clusters requiring urgent intervention due to suboptimal water injection patterns and other inefficiencies. The findings demonstrate significant disparities in oil production and water injection efficiency among the clusters, with one showing optimal performance and minimal maintenance needs. Incorporating these findings into the reservoir management activities does not only help in reducing unplanned outages but also optimizes resource allocation, enhancing overall field management by aligning operational strategies with real-time performance data.