Markov Chain Monte Carlo (MCMC) Parameter
Estimation for DCA Forecasting Model
Malik Nabeel Aamer
Oil and Gas Production Forecasting
● Uncertainty in production forecasting and reserve evaluation
● The decline curve could be smooth or noisy depending on factors such as well completion
quality and operating conditions
● The commonly used methods rely heavily on the experience of the evaluator and provide
little information on the uncertainty of the estimation
Importance of Accurate Production Forecasting
Accurate well production forecasts are needed for
● Operational decisions
● Long-term planning
● Commercial transactions
● Regulatory proceedings
Model Formulation: Arps Model
Arps Model, Empirical
Model
𝒒𝒕 = 𝒒𝒊(𝟏 + 𝒃𝑫𝒊 𝒕)
−
𝟏
𝒃
Proposed Distribution
𝒒𝒊,𝒑𝒓𝒐𝒑𝒐𝒔𝒆~ 𝑵 (𝒒𝒊, 𝟏)
𝑫𝒊,𝒑𝒓𝒐𝒑𝒐𝒔𝒆~ 𝑵 (𝑫𝒊, 𝟎. 𝟏)
𝒃 𝒑𝒓𝒐𝒑𝒐𝒔𝒆~ 𝑵 (𝒃,, 𝟎. 𝟏)
Parameters
Uniform Prior [0.1, 1,000,000]
𝑫𝒊 Uniform Prior [0.1, 50]
𝒃 Uniform Prior [0, 2]
Acceptance Ratio
𝜶 = 𝑴𝒊𝒏 𝟏,
𝒇(𝒒|𝜽 𝒑𝒓𝒐𝒑𝒐𝒔𝒂𝒍)
𝒇(𝒒|𝜽 𝒏−𝟏)
𝒒𝒊
non-informative prior distribution for each of the parameter normal distribution is proposed
Arps Model Mixing and Parameter Posterior Distribution
Forecast for 2nd year production ( Well 9) Overall production for all 153 wells
● Production prediction is heavily influenced by anomalies. Come up with a
solution in order to detect and remove such anomalies.
● Instead of Arps use more complex models. E.g. hierarchical models.
Future Direction

Using MCMC sampling technique for Well production forecasting

  • 1.
    Markov Chain MonteCarlo (MCMC) Parameter Estimation for DCA Forecasting Model Malik Nabeel Aamer
  • 2.
    Oil and GasProduction Forecasting ● Uncertainty in production forecasting and reserve evaluation ● The decline curve could be smooth or noisy depending on factors such as well completion quality and operating conditions ● The commonly used methods rely heavily on the experience of the evaluator and provide little information on the uncertainty of the estimation
  • 3.
    Importance of AccurateProduction Forecasting Accurate well production forecasts are needed for ● Operational decisions ● Long-term planning ● Commercial transactions ● Regulatory proceedings
  • 4.
    Model Formulation: ArpsModel Arps Model, Empirical Model 𝒒𝒕 = 𝒒𝒊(𝟏 + 𝒃𝑫𝒊 𝒕) − 𝟏 𝒃 Proposed Distribution 𝒒𝒊,𝒑𝒓𝒐𝒑𝒐𝒔𝒆~ 𝑵 (𝒒𝒊, 𝟏) 𝑫𝒊,𝒑𝒓𝒐𝒑𝒐𝒔𝒆~ 𝑵 (𝑫𝒊, 𝟎. 𝟏) 𝒃 𝒑𝒓𝒐𝒑𝒐𝒔𝒆~ 𝑵 (𝒃,, 𝟎. 𝟏) Parameters Uniform Prior [0.1, 1,000,000] 𝑫𝒊 Uniform Prior [0.1, 50] 𝒃 Uniform Prior [0, 2] Acceptance Ratio 𝜶 = 𝑴𝒊𝒏 𝟏, 𝒇(𝒒|𝜽 𝒑𝒓𝒐𝒑𝒐𝒔𝒂𝒍) 𝒇(𝒒|𝜽 𝒏−𝟏) 𝒒𝒊 non-informative prior distribution for each of the parameter normal distribution is proposed
  • 5.
    Arps Model Mixingand Parameter Posterior Distribution
  • 6.
    Forecast for 2ndyear production ( Well 9) Overall production for all 153 wells
  • 7.
    ● Production predictionis heavily influenced by anomalies. Come up with a solution in order to detect and remove such anomalies. ● Instead of Arps use more complex models. E.g. hierarchical models. Future Direction