Integrating Data Mining and Expert Knowledge for an Artificial Lift Advisory System

Abstract

Abstract This paper describes a new workflow to accelerate and improve decisions regarding where and when to apply an artificial lift system in fields with a considerable number of active wells. The workflow deploys a hybrid combination of a user-driven expert system and a data-driven knowledge-capturing system calibrated with historic data. Both systems interact to determine the right point in time to support a particular well with an artificial lift system. In the case presented, the mature gas field has a large number of operating wells, with predominantly manual operating data entry and a long processing time for newly acquired data. Due to the rapid decline rate in many wells, however, quick decisions are needed to improve productivity and hence the economics of each individual well. During the first two phases of the project, the asset team focused on data collection and workflow automation to speed up well production surveillance operations (e.g., gas rate calculation, estimation of critical velocity, etc.) (Mota, 2007). This paper documents the third phase, which addresses the knowledge-capturing and advisory components of the solution. Mature fields typically have significant field and asset expertise and a huge amount of historic data. Both information sources—the expert documentation and historic data—can be integrated to investigate past decisions and identify an optimum approach to field interventions. This paper describes the setup and implementation of a hybrid model that combines expert knowledge from asset engineers with the new knowledge discovered through the latest data-mining technology. The resulting system is then implemented in a fully automated workflow that identifies which wells require artificial lift. The results from a case study in a North Mexico gas field are presented. The reservoir is highly compartmentalized and requires fracturing as a way to increment well productivity. The data-mining approach used in this study is a special visualization technique, the self-organizing map (SOM), and a clustering algorithm (Zangl, 2003). The model was trained with historic production data, well test data, and information about historic well intervention decisions. In addition, expert knowledge from the asset engineers was introduced. The combination of data and expert knowledge enabled fast and reliable identification of the optimal time to install an artificial lift system to increase production while also effectively managing costs. This system reduces the typical decision time from several days to a matter of hours. The automated workflow runs immediately after the data is acquired and provides a continuous, up-to-date, and ranked list of proposed wells for artificial lift analysis. When new decisions are taken, the model can be updated for future use. The rapid analysis and decision cycle reduces lost production and improves overall field and asset value.