Use of Agent Structures for Event Detection: Identification of Wells Watering Out on Troll

Abstract

The industry-wide focus on Intelligent Energy in recent years has lead to more and better data being available, both onshore and offshore. Regrettably, processing tools have not seen a proportional improvement, and therefore engineers experience a data overload. Fairly simple alarms along with manual routine checks using independent, "offline" tools are dominating the daily tasks in production optimisation, and more intelligence and flexibility is thus needed. In this paper we suggest an agent structure allowing online, real-time event detection. In this work producing oil wells watering out on Trollwere used as an example. The detection algorithm was designed to search for significant drops in the wellhead pressure. By detection of such trends at an early stage, a successful correction and restabilisation is more likely. Introduction In recent years development in computer technology has lead to a considerable increase in availability, quality and quantity of data. More and more data is made available both onshore and offshore, and the acquisition frequency and storage possibilities have been improved considerably. This development has offered a new range of possibilities for improving operations, which is widely known as IO (Integrated Operations), e-operation or i-field. However, the handling capacity in the organisations has not been able to follow the vast improvement in data availability. A variety of analysis and visualisation programs exist, but data retrieval, preparation and configuration may be time-consuming and complex. This leads to unreleased potential, especially related to real-time and online applications. The amounts of data available and the possibilities this offers largely "overpowers" the capacities of the current tools and methods used, thus leaving what can be called a capacity gap between data availability and data handling. To keep up with the amounts of data, data-driven workflowscan be a useful tool, with a central element being automatic event-detection. By data-driven workflows we understand workflows where tasks, analyses and adjustments to be performed are triggered by events detected in the real-time data stream, as opposed to being part of predefined, scheduled routines. In certain contexts and areas this is partly done today, but many alarms and detections used today are not "smart" enough, resulting in both false detections and missed events. Although the equipment in the industry is becoming more automated, with less manual operations related to daily operation, there will still be tasks that may be too risky, too complex, or simply too expensive to automate, either due to the nature of the task or the state/properties of available equipment. The solution(s) presented here are mainly intended as a tool to help filling this gap, by assisting production/process monitoring and by highlighting irregularities. Nevertheless, a natural expansion of the system would involve more complex algorithms simulating more "intelligence", and could also include automation of tasks based on detections.