Applications of data analysis techniques for oil production prediction

https://doi.org/10.1016/j.engappai.2004.11.010Get rights and content

Abstract

This paper describes two data analysis techniques adopted in a Decision Support System (DSS) that aids users in predicting oil production of an infill well. The system generates predictions in the form of a possible range of cumulative production and length of production life of an infill well. The system also shows the worst and best case scenarios based on different production curves so that the expert can examine the plots of predicted production rates for each existing well and decide which model gives the best fit. The production curve of each individual well was mathematically modeled so that production values beyond the historical data can be produced. Decline curve estimation and neural network approaches were adopted for data analysis in the system. The system was tested with data from two groups of wells from two different fields in Saskatchewan, Canada. Observations on the suitable duration that the historical data set should cover and a comparison among different curve estimation and neural network models are presented.

Introduction

Petroleum engineers have searched for a simple but reliable way to predict oil production for a long time. Prediction of future production of petroleum wells is important for cost-effective operations of the petroleum industry. Production predictions can assist petroleum engineers in economic forecasts, and the approach often adopted by reservoir engineers is numerical simulation based on log and core analysis results. However, this process can be technically difficult, time consuming and expensive in terms of both labor and computational resources. At the same time, there is a vast amount of production data that is collected and stored, but hardly used. This set of production data can be utilized to build a model to predict production. Simple curve estimation could be adopted for this task but its accuracy is often suspect. Hence, the Artificial Neural Network (ANN) technique has been adopted as an alternative approach. The advantages of neural networks include its computational efficiency, non-linear characteristics, generation properties, and ease of working with high-dimensional data. This paper describes a Decision Support System (DSS) that incorporates both curve fitting and ANN approaches, and presents users with a range of possible solutions in the range of cumulative production and length of production of an infill well. By reviewing the range of possible solutions, the user can determine the best production forecast.

The objective of this study is to present this DSS for petroleum production. The basis of the prediction is the well's initial production and its economic cut-off. The assumption adopted in the investigation is that the production of an infill well is similar to existing wells in the same reservoir, therefore production curves of existing wells can provide useful information for future production of an infill well. However, it is also assumed that an expert user is best able to judge how relevant the information is from the existing wells. Hence, the system provides the best and worst case scenarios based on production curves from the existing wells, and the expert user selects among the plots of predicted production rates from each existing well to decide which model has the best fit. The system also generates a model based on the average production of the existing wells, which is provided to the user as reference.

In addition to generating the best and worst case scenarios of predicted production curves, the system also mathematically models the production curve of each individual well so that production values beyond the historical data can be produced. Both curve estimation and neural network approaches are adopted for mathematical modeling. In the curve estimation approach, the five types of linear, logarithmic, exponential, harmonic and general hyperbolic curve fitting methods were used. In addition, the system can also conduct neural network modeling and generate a model that associates the amount of oil produced per day in a certain month with the month index.

The paper is organized as follows. Section 2 reviews some background literature on traditional methods for predicting petroleum production, and these are primarily curve estimation and neural network approaches. Section 3 describes the methods and tools used to develop the DSS presented in this study. Section 4 describes the DSS. Section 5 provides some experiment results. Section 6 presents some observations and discussion, and Section 7 concludes the paper with summary and future work.

Section snippets

Petroleum production prediction

The five traditional methods of estimating, both physically and economically, remaining reserves of petroleum in an oil well include: (1) by analogy, (2) volumetrics, (3) material balance, (4) decline curve fitting, and (5) reservoir simulation (Thompson and Wright, 1985). Each of these methods can be applied independently and has its strengths and weaknesses. While all five methods can be used for predicting recoverable reserves of a reservoir, the methods have different data requirements. But

Methods and tools

The DSS adopts two basic modeling approaches: (1) estimation curve, and (2) neural network methods. They are presented as follows.

Components of the oil production predictor system

The DSS contains three main components: a user interface, history-matching models, and an analogue predictor. They will be discussed in detail as follows.

Experimental results

The Oil Production Predictor System was used for analyzing two groups of wells. Group 1 contains 14 short-life wells from the Oakly field and group 2 contains five long-life wells from the Midale field in Saskatchewan, Canada. Short-life wells last for less than 90 months while long-life wells last for 200 or more months.

Fitness of models

To investigate performance of the different methods on the training data, a comparison of the predicted values from all the methods versus actual production rate of one well at Midale was made. Among the curve estimation methods, hyperbolic and logarithmic curves display better fitting abilities. However, as can be observed in Fig. 4, the neural network model fits the training data better than the curve fitting methods.

However, as discussed in the previous section, a good fit to training data

Conclusion and future work

In recent years, considerable progress has been made in the development of methods for estimating future production of oil wells. However, these methods are usually either difficult to use or inaccurate. The DSS presented in this paper does not attempt to invent a novel prediction method. Instead, it is based on several simple available methods and gives users numerical and visual illustrations of the results so that experienced users can exercise their judgement and decide which scenario is

Acknowledgements

The authors would like to acknowledge the generous support of a Strategic Grant from the Natural Sciences and Engineering Research Council of Canada. We are grateful also for insightful discussions and support of Michael Monea of Petroleum Technology Research Center and Malcolm Wilson of University of Regina at various stages of the research.

References (12)

  • F. Aminzadeh et al.

    Estimation of reservoir parameter using a hybrid neural network

    Journal of Petroleum Science and Engineering

    (1999)
  • Baker, R.O., Spenceley, N.K., Guo, B., Schechter, D.S., 1998. Using an analytical decline model to characterize...
  • El-Banbi, A.H., Wattenbarger, R.A., 1996. Analysis of commingled tight gas reservoirs. SPE Annual Technical Conference...
  • M.J. Fetkovich

    Decline curve analysis using type curves

    Journal of Petroleum Technology

    (1980)
  • R.B. Gharbi et al.

    Universal neural-network-based model for estimating the PVT properties of crude oil systems

    Energy & Fuels

    (1999)
  • Z. Huang et al.

    Determination of porosity and permeability in reservoir Intervals by artificial neural network modeling, offshore eastern Canada

    Petroleum Geoscience

    (1997)
There are more references available in the full text version of this article.

Cited by (14)

  • Modelling two-phase Z factor of gas condensate reservoirs: Application of Artificial Intelligence (AI)

    2022, Journal of Petroleum Science and Engineering
    Citation Excerpt :

    Recently Saghafi and Arabloo (2018) used a genetic programming to propose a compressibility factor model for gas condensate reservoirs. Among many AI algorithms artificial neural networks (ANN) has gained immense popularity between the research community (Chamkalani et al., 2013; Chen et al., 2014; Gülüm et al., 2018; Jalali et al., 2007; Mahdaviara et al., 2020; Nguyen and Chan, 2005; Safari et al., 2014; Tatar et al., 2016; Zendehboudi et al., 2012; Zhang et al., 2019). This is due to the success of ANN in mapping inputs and outputs parameters in an easy, efficient and accurate way in compare to other AI algorithms.

  • A steam injection distribution optimization method for SAGD oil field using LSTM and dynamic programming

    2021, ISA Transactions
    Citation Excerpt :

    Existing time series prediction methods have great potential in oil production prediction problems, such as autoregressive integrated moving average (ARIMA) [8], artificial neural network (ANN) [9], extreme learning machine (ELM) [10], recurrent neural network (RNN) [11], and Elman neural network (ENN) [12]. However, ARIMA model is not able to process multiple input variables in time series prediction [13]. Traditional ANN model and ELM model cannot take account of the temporal relationships of steam injection and oil production data [13].

  • Optimization of tile manufacturing process using particle swarm optimization

    2011, Swarm and Evolutionary Computation
    Citation Excerpt :

    This ability to learn is perhaps one of the most valuable properties of an ANN because it enables a trained network to provide high-accuracy output for a set of previously unseen input data. Over years, countless applications of ANN have been reported in specialized literatures, some of which include Azamathulla et al. [21–23], where ANN was successfully employed to estimate the scour downstream of a ski-jump bucket, Nguyen and Chan [24] developed a decision support system (DSS) with neural network to predict oil production, Thillaivanan et al. [25] applied the ANN and Taguchi method to optimize the operating parameters of an electrical discharge machining (EDM) process, and Yazdi and Khorram [26], where the machining parameters of face milling process was modeled and optimized with ANN and response surface methodology (RSM) respectively. Despite its proven ability, ANN is considered in this study as a modeling tool where its implementation is discussed in the subsequent section.

  • Analysis of data for the carbon dioxide capture domain

    2011, Engineering Applications of Artificial Intelligence
    Citation Excerpt :

    Some relevant works on applying ANN modeling combined with sensitivity analysis (SA) have been conducted in the domains of business and engineering (Baker et al., 1999; Cullen and Frey, 1999; Embrechts et al., 2001; Poh et al., 1998; Fraedrich and Goldberg, 2000; Kleijnen, 1995). Nguyen and Chan (2005) also presented an ANN model inpredicting oil well productions. Sensitivity analysis (SA) is the study of how the variation or uncertainty in the output of a mathematical model can be apportioned, qualitatively or quantitatively, to different sources of variation in the input of a model (Cacuci et al., 2005).

View all citing articles on Scopus
View full text