A data mining approach to finding relationships between reservoir properties and oil production for CHOPS
Introduction
Cold heavy oil production with sand (CHOPS) is a primary oil extraction process for heavy crude oil that has been widely used in heavy oil production in the Western Canada Sedimentary Basin since the mid-1990s (Sawatzky et al., 2002). The intentional sand production along with the oil leads to the presence of wormholes, which are high-porosity, high-permeability zones in a reservoir (Mai et al., 2009). The generation and growth of wormholes are believed to enhance oil recovery (Chugh et al., 2000).
Some studies have been conducted to understand the factors affecting oil recovery for CHOPS. Most of them are based on reservoir simulation and focus on the modeling of wormhole networks. Yuan et al. (1999) constructed reservoir engineering tools to assess cold production footprints with history data matching. Liu and Zhao (2005) used a fractal network, relying on a probabilistic active walker approach, to model wormhole networks; however, sand production rates were required to generate the network.
Tremblay (2009) investigated a series of multilateral wells in the modeling of individual wormholes. Rivero et al. (2010) employed an equivalent damaged zone to model the extent of growth of a wormhole network. Istchenko and Gates (2012) proposed the use of the results of history matches versus field data to adjust model parameters. The history-matched model was then applied to predict production trends of CHOPS field operations.
Time-lapse seismic imaging has been proposed for the monitoring of the production performance over the time, but most reservoirs are thin with less than 10 m of net pay (i.e., thickness of rock that contains economically producible oil), which is a big challenge for the use of this method (Sawatzky et al., 2002).
With the existing wormhole models, it is difficult to precisely evaluate and describe the factors affecting oil recovery for CHOPS, due to the complex variation of geological conditions. Although raw data have been accumulated over the years, the challenge is often to marshal and interpret an overwhelming amount of raw data to discover the relationships between reservoir properties and oil production.
To achieve this goal, a different approach is taken in this paper to solve the problem. Instead of studying wormhole generation and growth, we propose to apply data mining methods on large CHOPS datasets to find relationships between the reservoir properties and oil production. The data mining approach includes three steps, as shown in Fig. 1 and listed in the following:
- i
Data preparation. Field data are huge and noisy. The proper data reflecting reservoir properties need to be acquired. Reservoir properties are identified to describe reservoir characteristics. In addition to some common parameters, such as porosity and permeability, two new parameters – a fluid mobility factor and the maximum inscribed rectangular of net pay (MIRNP) – are proposed. Additionally, three parameters to represent the production performance of wells are proposed: the peak value, effective life cycle and effective yield.
- ii
Data filtration. The extracted data need to filter further before data mining. In this step, the fuzzy ranking method is used to rank the importance of the identified reservoir properties in terms of oil production; and, appropriate reservoir properties are chosen.
- iii
Data mining. Association rule mining is used to identify potential patterns between the reservoir properties and the production performance.
In this paper, 118 wells were selected from the Sparky Formation in the Lloydminster heavy oil field in Alberta, as a case study to illustrate the approach.
The rest of the paper is organized as follows: Section 2 describes the proposed integrated data mining methodology for CHOPS fields. In Section 3, a case study of a Lloydminster heavy oil field in Alberta is presented step by step. Section 4 provides the conclusions and discussions.
Section snippets
Integrated data mining approach for CHOPS
In this section, we propose an integrated data mining approach for CHOPS. The approach is explained by following the three steps shown in Fig. 1, i.e., data preparation, data filtration, and data mining.
Case study
In this study, data from CHOPS wells in the Lloydminster heavy oil area were collected from Divestco data (2014), as shown in Fig. 5. The reservoir data for this study were comprised of well logging data and core data. Other factors, such as the formation pressure, also impact production. However, due to the lack of availability of other datasets for the area, in this paper we only focus on the relationships between reservoir properties and production yields. We also assume the same perforation
Conclusions and discussions
This paper proposes a data mining approach to finding the relationships between reservoir properties and cold heavy oil production with sand (CHOPS) performance. The Sparky Formation in the Lloydminster heavy oil field in Alberta of Canada was used as a case study to illustrate the approach. On the basis of a petrophysical data analysis, a set of parameters that can characterize the net pay was extracted.
It was determined that the cumulative fluid mobility factor (KRc) is an important parameter
Acknowledgment
The research is supported by the Natural Sciences and Engineering Research Council of Canada Discovery (Grant No. 355996-2013) Grant to the second author and Petroleum Technology Research Center Sustainable Technologies (Grant No. HO-UOC-00004-2012) for Energy Production Systems program.
We would like to thank Divestco Ltd. for providing data for the studied wells. We also acknowledge our group members Xiaodong Sun and Xi Wang for their technical support and Jihong Luo and Weifeng You for the
References (24)
- et al.
Product portfolio identification based on association rule mining
Comput.-Aided Des.
(2005) - et al.
Input variable identification – fuzzy curves and fuzzy surfaces
Fuzzy Sets Syst.
(1996) - et al.
Searching customer patterns of mobile service using clustering and quantitative association rule
Expert Syst. Appl.
(2008) - et al.
A hybrid framework for reservoir characterization using fuzzy ranking and an artificial neural network
Comput. Geosci.
(2013) - et al.
Smart oilfield data mining for reservoir analysis
Int. J. Eng. Technol.
(2010) - et al.
Mainstream options for heavy oil: part I—cold production
J. Can. Pet. Technol.
(2000) - et al.
Mining gene expression databases for association rules
Bioinformatics
(2003) - Divestco data. 〈http://www.divestco.com/getdoc/14af7505-8813-406e-b8b7-757f9a751ff5/Add-On-Datasets〉.aspx (accessed...
- Dresser Atlas, 1979. Dresser Atlas Log Interpretation Charts, Dresser Atlas, Dresser...
- et al.
Rxo/rt methods for wellsite interpretation
Log Anal
(1972)
Data mining concepts and techniques
Cited by (7)
A new method of well clustering and association rule mining
2022, Journal of Petroleum Science and EngineeringCitation Excerpt :Moreover, it is possible to use these approaches to forecast oil and gas production in both unconventional and conventional reservoirs (Mohaghegh, 2017; Saikia et al., 2019; Cao et al., 2016) and to predict the performance of vertical and horizontal production wells (Alzahabi et al., 2020). In this approach, various reservoir data, i.e., temperature, pressure, permeability, well data, hydraulic fracture data, Young's modulus, and Poisson's modulus, are used to predict model parameters (Li and Han, 2017; Cai et al., 2014). Furthermore, machine learning algorithms are implemented in enhanced oil recovery (EOR) screening studies (Khazali et al., 2019) or the design of water-alternating carbon dioxide injection in the laboratory (You et al., 2020).
Model development for shear sonic velocity using geophysical log data: Sensitivity analysis and statistical assessment
2021, Journal of Natural Gas Science and EngineeringCitation Excerpt :The soft computing approaches (or connectionist models) include artificial neural network (ANN), recurrent neural network (RNN), adaptive neuro-fuzzy inference system (ANFIS), convolution neural network (CNN), extreme learning machine (ELM), gene expression programming (GEP), support vector machine (SVM), least squares support vector machine (LSSVM), and hybrid smart models (Arabloo et al., 2015; Ghiasi et al., 2015; Kamari, 2015; Kamari 2014a; 2014b; Zendehboudi et al., 2018). These tools are widely employed to construct log data-driven models for estimation of reservoir properties and rock mechanical properties related to reservoir characterization, geomechanics, and drilling engineering fields (Nikravesh and Aminzadeh, 2003; Hsieh et al., 2009; Cranganu et al., 2010; Asoodeh and Bagheripour, 2012; Cai et al., 2014; Ceryan, 2014; Arabloo and Rafiee-Taghanaki, 2014; Anifowose et al., 2014a, 2014b, 2015; Akande et al., 2015a, 2015b; Amiri et al., 2015; Esfahani et al., 2015; Ebrahimi and Khamehchi, 2016; Zoveidavianpoor, 2017; Ismail et al., 2017; Wang and Peng, 2019; Anemangely et al., 2019; Sudakov et al., 2019; Miah et al., 2019, 2020a; Movahhed et al., 2019; Dumke, 2019). These connectionist tools are known to be used for various prediction and classification purposes with low costs.
A web-based visual and analytical geographical information system for oil and gas data
2017, ISPRS International Journal of Geo-InformationGeovisualization for association rule mining in oil and GasWell data
2017, ISPRS International Journal of Geo-Information