Nazarenko, Maksim (2016-05). Probabilistic Production Forecasting and Reserves Estimation in Waterflooded Oil Reservoirs. Master's Thesis. Thesis uri icon

abstract

  • The importance of uncertainty quantification and risks assessment in the petroleum industry cannot be overstated. Uncertainty will always be present in production forecasts and reserves estimates. Underestimation of uncertainty when estimating reserves and profitability of projects can lead to poor decision making and disappointment. Water Displacement Curve (WDC) models allow engineers to estimate reserves and forecast production performance in waterflooded oil reservoirs taking into account either liquid or water production. Compared with Decline Curve Analysis (DCA), WDC models are expected to perform better in forecasting oil production in waterflooded oil fields. In this study I applied Bayesian methodology and Markov Chain Monte Carlo (MCMC) methods with WDC models. I also developed a Multimodel approach based on eleven WDC models to quantify uncertainty in production forecasts by assessing differences in matches and forecasts provided by each model. Both Multimodel and MCMC with WDC models were calibrated and compared to MCMC with DCA methods. Reliability of the developed methods was assessed using production history of 100 wells from actual waterflooded oil fields. I performed hindcast studies in which I assumed that some fraction of the actual historical production data is known (6, 12, 24 and 36 months) and the rest of the actual production is unknown (5 - 7 years). I then matched the assumed known production fraction of the history and forecasted production to the end of the actual historical period. The cumulative production at the end of the hindcast is compared to the actual cumulative production at this time to test the probabilistic reliability of the methodology when production history is limited. The study showed that o WDC Multimodel, MCMC with WDC and MCMC with DCA are well-calibrated probabilistic methods o WDC Multimodel performs more than 20 times faster than MCMC with WDC and MCMC with DCA techniques having the same level of reliability o Compared with MCMC using DCA, WDC Multimodel and MCMC with WDC show more reliable results when the history matching period is less than 24 months Computer software was developed during this research to make the process of calculations more convenient.
  • The importance of uncertainty quantification and risks assessment in the petroleum industry cannot be overstated. Uncertainty will always be present in production forecasts and reserves estimates. Underestimation of uncertainty when estimating reserves and profitability of projects can lead to poor decision making and disappointment.

    Water Displacement Curve (WDC) models allow engineers to estimate reserves and forecast production performance in waterflooded oil reservoirs taking into account either liquid or water production. Compared with Decline Curve Analysis (DCA), WDC models are expected to perform better in forecasting oil production in waterflooded oil fields.

    In this study I applied Bayesian methodology and Markov Chain Monte Carlo (MCMC) methods with WDC models. I also developed a Multimodel approach based on eleven WDC models to quantify uncertainty in production forecasts by assessing differences in matches and forecasts provided by each model.

    Both Multimodel and MCMC with WDC models were calibrated and compared to MCMC with DCA methods. Reliability of the developed methods was assessed using production history of 100 wells from actual waterflooded oil fields. I performed hindcast studies in which I assumed that some fraction of the actual historical production data is known (6, 12, 24 and 36 months) and the rest of the actual production is unknown (5 - 7 years). I then matched the assumed known production fraction of the history and forecasted production to the end of the actual historical period. The cumulative production at the end of the hindcast is compared to the actual cumulative production at this time to test the probabilistic reliability of the methodology when production history is limited.

    The study showed that
    o WDC Multimodel, MCMC with WDC and MCMC with DCA are well-calibrated probabilistic methods
    o WDC Multimodel performs more than 20 times faster than MCMC with WDC and MCMC with DCA techniques having the same level of reliability
    o Compared with MCMC using DCA, WDC Multimodel and MCMC with WDC show more reliable results when the history matching period is less than 24 months

    Computer software was developed during this research to make the process of calculations more convenient.

ETD Chair

publication date

  • May 2016