Good Decision Tree Analysis In Petroleum Industry Literature Review Example
There is a high risk involved in the exploration and production of petroleum and other hydrocarbons. The high risk can be attributed to the uncertainty of geological concepts such as reservoir seal, geological structure and hydrocarbon charge. Also, there are economic evaluation uncertainties such as those related to costs, probability of striking reservoirs that are economically viable, oil price uncertainties etc. The engineering parameters contain a high degree of uncertainty even at the production stage. The critical variables that pose uncertainties in development and production stage are; the production schedule, the oil quality, the cost of operations, the characteristics of the reservoir etc.
These uncertainties make decision making in the petroleum industry to be a high-risk business. Managers in this industry have to continuously make decisions on how to allocate the scarce and competing resources among projects that have a substantial financial and geological risk. To make sound decisions corporate managers could employ data modeling approaches e.g. the neural-decision tree model. This model considers the correlations or dependencies of the variables or attributes, thus it is a distinctive model.
In the recent years, artificial intelligence has increasingly been accepted in the petroleum industry. The nonlinear prediction methods in the oil and gas industry are now using artificial neural network (ANN) models (Li & Chan, 2010). The ANN models however generate “black boxes”, thus not helping the decision makers comprehend the exact relationships of the variables being modeled (Li & Chan, 2010).
In its simplest form a decision tree is simply an idea generation tool that refers to a model of decisions with their possible outcomes, including the probability of the event outcomes, the cost of resources to deliver the outcome and the utility of the each outcome. A decision tree represents a predictive model in data mining. Decision trees have a number of advantages such as that of being easy to understand and interpret when used by a petroleum engineer. Also, a decision tree is easily and quickly constructed as compared to other methods.
In petroleum engineering decision trees can be used to classify permeability predictions data based on logs from various wells (Perez et al. 2005). Also, decision tree analysis can be used to estimate the uncertainty range in production prognosis of reservoirs (Jensen, 1998). Similarly, decision tree models can be used together with fuzzy models to help in ranking of various reservoirs depending on the quantity of gas production for each (Agbon et al. 2003).
Decision trees in simplest form assume some level of independence of the available data. However, in petroleum engineering, the data sets available usually embody attributes with some interdependence. Thus, it is necessary to allow for possible dependencies in the attributes of the given data when using decision trees. The neural-decision tree model allow for this modification.
The Neural-decision tree model.
The neural-decision tree (NDT) model makes use of both the decision tree algorithms and the neural networks. The fundamental strength of this model is its ability to capture attribute dependencies, thus, it is more accurate in decision making as opposed to decision trees that do not take into consideration the dependencies among the attributes. The figure below shows the NDT model assuming a data set that is of mixed-type.
For the NDT model the rule extraction steps as illustrated in the above fig 1 are as follows;
The mixed-type dataset is divided into two parts of nominal subset and numerical subset. If the dataset is of pure-type, such division is not necessary.
Using the numerical subset, a feed-forward back-propagation train is employed and the weights collected between the first hidden layer and the input layer. Based on the generated weights, change each attribute value accordingly.
Using the categorical subset, a back-propagation train is employed and depending on the weights generated the categorical attributes are classified.
The new numerical subset is combined with the new categorical subset to come up with a new dataset that is numerical-categorical-mixed.
This new data set becomes the input to C$.5 system, which then gives as its output a decision tree and together with its rules.
An evaluation is conducted on the results to ascertain a reliable and accurate estimate of the predictor or classifier model.
Application of Decision trees in the Oil and gas industry.
The NDT has been can be used to predict petroleum production from a given well. To mitigate the financial risk caused by uncertainty in well drilling, proper economic evaluation has to be carried out on any new oil well before a decision to drill is made. The evaluation is made on the basis of the prediction of the well’s production. Due to the complexity of the subsurface conditions of any given reservoirs, challenges arise in accurately predicting any well’s production. The standard approach used in the oil and gas industry is use of curve fitting-fitting methodologies or use of reservoir simulations, all of which are complex and time consuming. However, use of the decision NDT model can simplify this process and give a good prediction on the viability of a well before it is drilled. In these applications NDT model is principally used to identify relationships among the main variables used to predict oil production. The performance of this model is highly dependent on the accuracy of the data sets used. This method was used to predict petroleum production in a number of wells in Saskatchewan, Canada.
Decision trees can also be used in the petroleum industry to determine the ideal artificial lift method to be used. In this case the first decision step is to limit the possible outcomes based on the technical specifications. After that the secondary technical problems are considered such as temperature constraints. Thirdly investigations are conducted to establish the technical feasibility of the remaining methods. However, this process is not only complex but also subjective. Thus different corporation managers are likely to arrive at different systems even when given the same parameters.
Decision trees have also been used in Nigeria to determine the ideal mud system. There are several mud systems available in the market to drill a well; with good decision analysis using the decision trees the cost-effectiveness of the available options becomes apparent.
Li, X., & Chan, C. (2010). Application of an enhanced decision tree learning approach for prediction of petroleum production. Engineering applications of artificial intelligence, (23), 102-109.
Agbon, I.S., Aldana, G.J., Araque, J.C., 2003. Fuzzy ranking of gas exploitation opportunities in mature oil fields in Eastern Venezula. SPE paper 84337 presented at SPE Annual Technical Conference and Exhibition, Denver, CO, USA, 5–8 October 2003
Jensen, T.B., 1998. Estimation of production forecast uncertainty for a mature production license, SPE Annual Technical Conference and Exhibitions, New Orleans, USA, SPE 49091, September 1998.
Perez, H., Datta-Gupta, A., Misra, S., 2005. The role of electrofacies, lithofacies and hydraulic flow units in permeability predictions from well logs: a comparative analysis using classification trees. SPE Reservoir Eng. Eval. 8 (2) April.