2. Six Sigma Prediction Model
The main objective is to find an optimal value for each well-type that will minimize the deviations from actual value. This optimal value depends on historical data, underestimation and overestimation penalties, and other factors. The best model for this scenario is “Newsvendor” model.
2.1 Factorstoconsider
To be practical, uncertainties cannot be predicted by a single scenario. However, decision maker should make an informed judgment and hope for the best. Therefore, the estimation model should incorporate as many factors as possible that help shape these uncertainties; the historical data can give an indication about the most likely value for both cost and duration. The historical data can give the following information: (1) Actual cost and duration , (2) Learning experience, (3) Sources of variation within year or between years, and how much other factors such as tools or drilling methodology change contribute to the variation in the data. Relying solely on historical data to predict for future might mislead the prediction model. Thus, having expert’s belief or opinion about what should be actual cost and duration is essential. The challenge is how to help expert set a prior distribution for actual cost. The learning experience is modeled using Wright (1936) model.
Expert opinion about actual cost and duration will benefit the model and make it more practical. The reason is that the historical data may have an average accuracy level. This “average” accuracy level could be a result of missing data, unclear method for collecting data, or that data are gathered from various departments in oil and gas companies which may have unclear measurement system. To help experts, the model extracted relevant information (as summarized previously). These relevant pieces of information include: (1) Trends in the data, (2) Sources of variation, and (3) Learning experience. The end result of the expert opinion is a probability density function that gets updated using the historical data. However, this is not rigid and experts can still have an influence in the end result distribution as will come in the example provided in case-study section. Moreover, the expert state of knowledge can be explained by a member of the beta family distribution. It is entirely possible that the state of knowledge will not be adequately described by a beta distribution, but the flexibility that beta family distribution has makes it “effectively” capture expert opinion.