The objective of this study was to assess the feasibility of application of analytics techniques in a new heavy oil asset in Kuwait in the following areas: data integration and visualization to support Well, Reservoir and Facility Management (WRFM), understanding well production behavior and their link to reservoir parameters, investigating reasons for sand failure. The study also aimed to highlight focus areas that would facilitate the full field implementation of analytics as a viable WRFM tool. Due to the green field nature, the current volume of data is relatively small (versus mature assets), and not all the data required is available yet. The study started with formatting, processing and integrating all the available field data into a data management tool. Integrated visualizations were tested to detect early trends of sand and water production. Machine Learning algorithms such as Random Forest, Decision Tree and Neural Network were applied next to the sand prediction problem, focusing on identifying root causes for repetitive sand failures in wells and if possible to predict initial or subsequent sand failures. The study indicated that integrated visualizations are very promising in support of WRFM in this field in the short term. Early signs of water breakthrough were correlated, particularly in wells with specific combinations of geological features and completion strategies. The sand prediction problem proved to be very challenging to the Machine Learning approaches, with limited success in predicting the historical occurrences. The results indicated that these techniques are likely to be more applicable once the volume of data increases, particularly the higher resolution data (real time data from artificial lift equipment), as well as by incorporating additional data types (sand production measurements during tests, which require additional resources to execute) and other data types available but not tested yet (log data derived parameters). Overall results indicated that analytics should be strongly considered as a valuable tool in the short to medium term in this new field, with efforts in data acquisition and management of the data types identified. Most of the existing publications on this topic are related to analytics applied in mature fields. This study, conducted in a field still in the early stages, showed the value for early implementation, and highlights that early planning for focused data acquisition can facilitate building initial data-driven models which can be used for prediction purposes. The paper is expected to provide a valuable reference for new heavy oil projects, either under definition or in early production, for the application of data analytics and extend learnings based on machine learning.
Production forecasts provide fundamental input to upstream business decisions, for example in resource volume reporting, field development and production planning. Very often in mature fields, having sufficient cumulative production, field forecasts are performed based on aggregating type curves for new wells and decline curve analysis for existing wells. Conventionally, they are derived by manual trend fitting which is subjective and any iteration incorporating new data is time consuming. The advanced analytics approach presented in this paper can provide rapid and credible forecasts along with more robust quantification of uncertainties as compared to the conventional manual approach. The first step involved in the forecast automation using advanced analytics is data integration in which the production, geological, and surveillance data are combined. Then integrated dashboards are created to infer trends in production behavior. Based on these trends at the most granular forecast level, petroleum engineering based algorithms are developed to automatically select the decline period. Thereafter, multiple scenarios of historical data are generated based on historical allocation and measurement uncertainty. Through each set of these historical scenarios, best fits curves following Arp's equation are extrapolated into the future to generate multiple forecast scenarios. Finally, field level forecast is generated by probabilistically aggregating individual granular level forecasts. The automated forecast program developed is computationally very fast as 500 forecast scenarios at the well level can be generated in 30 seconds, while the full field forecast with 50 wells takes about 30 minutes to generate. Automation of decline period selection and curve fitting eliminates the subjectivity by standardizing the process and reduces the chances of manual errors. Abandonment criteria, discounting and uptime variations can easily be accommodated in the automated process. Visualizations are utilized at each step for quality check and analysis. Forecasts from alternate methodologies are used to validate the forecast ranges coming out from this method. Uncertainties quantification in this approach is found to be more quantifiable and consistent compared to the conventional deterministic approach. Production dashboards created in this workflow by integrating production, surveillance, geological data and forecasts are a very effective tool to perform field reviews and communicate outcomes. The approach described above for decline curve analysis can easily be extended to any type-curve based forecasting. Automation of performance-based decline curve and type curve forecast methodologies has the potential to reduce huge manhours involved in their periodic updates by an order of magnitude that can be utilized to carry out other critical analysis. It is very useful in mature assets with large well inventory, huge dataset and where continuously new data are being added. Standardization of workflow, implementation ease and accuracy will tempt practitioners to use it and thereby develop skills in data analytics.
Having less data is challenging as it adds to the uncertainty around our understanding. On the contrary, huge volume of existing data with continuous additions poses challenges to incorporate them into subsurface models. Data analytics involving integration of large volume of geology, production and surveillance data at a single platform, analysis of data and data trends depicting a physical process can be used to come up at technically robust results in a short time. Data driven analysis and forecasting is carried out for a mature steamflood. Firstly, all data including production, injection, pressure, well test, well, static models, petrophysical logs, GIS data are integrated on a single data analytics platform. Trends in data are established and visualized based on the production, injection and geology enabling production performance analysis at various aggregated levels. The established trends are studied to understand the effectiveness of steam injection recovery mechanism, interference and are then forecasted. Data like PLT and static model information earlier analyzed in silos are integrated and now examined together with production data. Due to data integration at a single platform, connecting and visualizing time series data with spatial location to arrive at certain aggregation or cluster is now rapid and credible. Observations and outcomes from data driven analysis correlate well with the recent years of historical data for well groups with mature production and injection history. For well groups with new and upcoming wells, data driven method may get more biased towards existing but sparse performance thus not capturing the full uncertainty in forecast. Data driven analysis method becomes effective when sufficient data is available to infer trends and results that can be validated with alternate estimation or new data. This technique finds wide application in brownfields with huge data that are either not analyzed or integrated with other information to develop a holistic understanding of subsurface uncertainties. Aided with effective visualization, this technique provides an alternative methodology to enable rapid decision making.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.