Universidad de los Andes, Bogotá, Colombia
Universidad de los Andes, Bogotá, Colombia
Universidad de los Andes, Bogotá, Colombia
ABSTRACT: This paper presents the experience of a conjoint work between UniAndes and BP-Colombia, consisting in a training-consulting program in decision analysis for solving BP decision problems of high economic impact, related to technologies selection, oil exploitation strategies, and reserves estimation probabilistic models. Around 60 engineers have been trained in the use of Decision Trees, I. Diagrams, Bayesian Networks and AHP. The program has had a remarkable impact on BP decision making culture.
PALABRAS CLAVE: Decision Analysis, Risk Analysis, Montecarlo Simulation, Analytic Hierarchy Process, Influence Diagram, Decision Tree, Bayesian Network.
This article describes an experience developed jointly between the University of Los Andes and BP Exploration Company – Colombia, consisting of a Decision Analysis training/consulting program aimed at structuring and solving decision-making problems bearing a high-impact risk for BP Colombia. Various problems related to the selection of drilling technologies, exploitation strategies, and probabilistic models to estimate reserves and wellwork costs have been formulated and solved, using the general decision analysis approach and the most typical tools and models of this discipline.
This article is structured as follows: 1) Introduction; 2) Conceptual Framework, which describes the conceptual approach used by the consultants, the general methodology used to structure problems related to decisions under risk conditions, and a list of some books and articles that has served as reference to develop the project; 3) General description of the training/consulting program, including the way that such University-Company program was conceived and structured, and a summary of the main types of decision-making problems that BP Colombia must cope with under risk situations and of why the methodologies, models and tools of the decision analysis are particularly relevant; 4) Detailed description of some of the problems that have been the object of consulting and the specific methodologies and models used to solve them; 5) Results, where we summarize the most relevant results produced by the project, both for the company’s various areas involved and for the company’s corporate culture.
1.2. CONCEPTUAL FRAMEWORK
Through the various experiences that the researchers have had in structuring and solving decision-making problems in Colombian companies, they have verified that the conceptual aspect and the rigor in the application of the decision-analysis methodologies are as important as identifying the models that best fit the solution of the problem. It is certainly not easy to persuade the companies, always pressed for time, that in order to successfully solve a decision-making problem, it is necessary to carefully pursue a series of steps leading to transforming a problem situation, sometimes vague and confusing, into a structured problem whose different components and variables are clearly identified and described in detail.
The ultimate objective of Decision Analysis is to improve the quality of companies’ decision-making process. According to Matheson and Matheson , it is necessary to have an adequate conceptual framework, generate creative and feasible alternatives, have relevant and reliable information, have clear decision-making criteria, use a correct logical reasoning (appropriate decision-making models) and, finally, be committed to act.
Quantitative models and tools are not enough by themselves to solve decision-making problems involving a certain level of complexity. Before deciding on the type of models and tools to use in the analysis, it is necessary to accurately describe the decision-making problem and structure it appropriately based on its main components, the relevant variables, the actors or agents involved in the decision-making process, and the available alternatives, among other aspects.
The methodology described below is aimed precisely at that purpose, and has been conceived by Castillo , based on his research and consulting work in this field. This methodology has been used successfully in various research and consulting projects , in a variety of decision-making problems pertaining to Colombian organizations including oil companies, financial institutions, telecommunication companies and universities.
Figure 1 summarizes the main steps of the methodology, which must be adjusted and modified according to the specific characteristics of the problem under analysis. The methodology has been conceived as a general and flexible guideline that must under no circumstance become a straight jacket when structuring the decision-making problem.
Figure 1: Decision Analysis Methodology
When comparing this methodology with other Decision Analysis methodologies developed and published by important authors such as Clemen  and von Winterfeldt , we can conclude that these three methodologies have similar structures and are applicable to a broad range of decision-making problems, regardless of whether the models used are typical Decision Analysis models or not. The methodology developed by Castillo clearly explains each of the stages of the decision-making process, and is therefore easily applicable to structure and solve various types of complex actual problems. Clemen’s methodology does not emphasize problem structuring so much, but it is very useful to make a detailed analysis of the alternatives and of the results of the evaluation. Von Winterfeldt and Edwards’ methodology is very powerful to define and structure the problem, and thus it would be a very good supplement for any of the other two.
Description of the Methodology
Description of the Situation
This stage consists of preparing a synthesis of the problem and of its main components. It’s as if the analyst had to write a one-page article on the problem in an academic magazine. Such article must summarize the main aspects of the problem, the relevant variables to be analyzed, the possible solution alternatives and the possible approaches to find a solution.
This stage involves the following steps:
Expressed problem situation: One must identify and describe the main aspects of the problem, the main actors or agents involved, and the most important relationships among them. Usually, the relationships among actors must be expressed through variables. This includes differentiating decision variables from random variables or sources of uncertainty, when analyzing the problem.
Generation of alternatives: Based on the analysis produced during the previous step, one must describe the decision alternatives (or solution alternatives) of the problem. Normally, it is necessary to consult with the decision makers and/or the people that are most familiar with the problem under analysis, in order to generate the decision alternatives that will be taken into account in the decision-making model.
Scope of the analysis (objective formulation): In this step, one must clearly state the problem to be solved, the objectives to be reached and the possible limitations of the analysis.
Identification of models and tools: Based on the prior analysis, and after reading and reviewing the literature available on the problem under analysis, one must identify precisely the models and tools to be used to evaluate the various decision alternatives. This step is very important for the analysis process, since this is where we define the type of information required to analyze the problem and to build the decision-making model(s). At this point we must know what type of model (or models) we will use and the possible order in which we will use them. Normally we choose one or several models such as Decision Trees, Influence Diagrams, Bayesian Networks, Analytic Hierarchy Process, Optimization and Simulation, among others. Sometimes, rather than designing a precise model, what we must design is a specific methodology to analyze the problem, which may include several steps and involve various models and analysis tools.
Obtaining the information
During this stage of the methodology, based on the identification of relevant variables and the selection of the models to be used, we must collect all the information and the data required to analyze the problem and to populate the models.
Model formulation and construction
Conceptual structuring of the model: The idea here is to place the problem in a conceptual model, adequately representing the decision-making process, its relevant variables, the decision-making criteria and the alternatives to be evaluated. A good conceptual model is the fundamental basis of a good quantitative model. As mentioned above, many decision-making problems require developing a relatively complex model or methodology, which will in turn require the use of different types of models. In this latter case, the conceptual model must accurately reflect the sequence in which the various models will be used, and their role during the various stages of the analysis.
Detailed construction of the model and description of its variables: This is the detailed construction of the model, incorporating the various relevant variables and the decision alternatives, usually resorting to the use of specialized decision-making software such as DPL, Expert Choice, Hugin and Crystal Ball, or optimization, simulation and statistical analysis programs.
Conceptual interpretation of the model: In decision analysis problems, it is key for the decision-evaluation models to have a clear conceptual interpretation of the persons involved in the problem. It is necessary to understand that decision-making models are above all models designed to serve as support for decision-making. Thus, in practice, it makes no sense to develop cryptic models that select alternatives but which do not allow for an easy interaction between the analysts and the decision makers.
Validation of the model: It may seem a commonplace, but it is very important to verify whether the model is working in the right direction, whether it allows for an appropriate formulation of the logic behind the decision-making process, and whether it incorporates the most important variables of the problem.
Obtaining and analyzing results
Selecting the best alternatives: Once the final model is satisfactory because it adequately represents the decision-making problem, we proceed to produce the results of the evaluation, identifying the best decision alternative(s) or the best solution(s) to the problem.
Analysis of final alternatives: Once the alternatives are evaluated, we must use the results produced by the model to identify the strengths and weaknesses of each alternative.
Additional Clarifications on the Use of the Methodology
Complexity of the reality: Organizations’ actual problems are normally highly complex, due to the huge amount of aspects involved, the diversity of the relationships among the actors, the multiple objectives that must be covered simultaneously, the specific characteristics of the decision-making team and the little availability of relevant information, among other things. So in practice, the models representing decision-making processes are usually a gross simplification of reality. When structuring and analyzing a decision-making problem, it is important to seek a compromise between building a reasonable simple model and adequately representing the problem. Exaggerating the level of detail when designing a model can lead to the impossibility to complete its design, usually due to the difficulty to specify the variables or the relationships among them, or to the impossibility to obtain excessively detailed information. On the other hand, an over-simplification of the problem that must be analyzed may lead to building models that describe the problem very poorly and, in this case, they are not a good support for decision-making. In summary, a good model is that which is capable of incorporating the most relevant aspects and variables of the problem, satisfactorily representing the decision-making process and its logical sequence, and producing results capable of becoming an actual support for the decision-making process.
The most important recommendation regarding this aspect is to give initial priority to the global and strategic aspects of the problem, instead of letting the details overwhelm us. The author’s experience confirms the thesis that most people tend to get entangled in the details of the problems and find it difficult to conceptualize and think strategically.
Analysis dynamics: In spite of the fact that the various stages of the proposed methodology have been presented on a sequential basis, in reality, the structuring of the problem and the construction of the models are dynamic in nature, and it is often necessary to go back to a prior stage due to problems identified at a subsequent stage. Frequently, for example, there is no information on some variable, and it is then necessary to go back and review the list of relevant variables and, eventually, the types of models that can be used for the analysis. Likewise, it may happen that while doing the detailed construction of the model, one may find the need to use some variables that were initially not specified. It is worth emphasizing that, according to the author’s experience in the construction of decision-making models, any flaws affecting stage 2 of the methodology (structuring of the problem) may lead to serious and expensive problems when building the decision-making models.
Besides the general methodology described above, the authors of this article have found permanent guide for their training/consulting program in the books by Clemen  and Matheson and Matheson , and Saaty , among others. In specific subjects such as Bayesian networks, the methodologies and models proposed by Nadkarni  and Neapolitan  have been particularly useful. For subjects directly related to oil reserve estimates and other subjects related to oil exploration and exploitation, the work by Capen , Fetkovich , Rose  and Smith  has been particularly useful.
1.3. GENERAL DESCRIPTION OF THE TRAINING/CONSULTING PROGRAM
The training/consulting program conceived by the group of researchers of the Los Andes University for BP Colombia comprises four phases. During Phase 1, the University consultants held several work meetings with BP engineers, with the purpose of jointly identifying relatively simple decision-making problems, which, once structured and solved, could be used as real company cases to illustrate the various methodologies and models covered by the training program. This first phase has turned out to be key to motivate the engineers before and during their training on the practical relevance of the use of models for decision-making under risk conditions.
Phase 2 is an intensive training program of a little over 30 hours, aimed at familiarizing the participants with the fundamental notions of Decision Analysis under Uncertainty and Risk Analysis, covering the main concepts of decision analysis and the methodology described under section 2 hereof. This training also includes in detail the main decision-making models such as Decision Trees, Influence Diagrams, Bayesian Networks, Analytic Hierarchy Process and Montecarlo Simulation. For each model, workshops are conducted in a computer room where the participants structure and solve decision-making problems using specific computer decision-making software including DPL, Hugin, Crystal Ball and Expert Choice.
Phase 3 takes place upon completion of Phase 2, and it consists of a series of consulting meetings where the group of University consultants talk about structuring the company’s specific decision-making problems, using methodologies and models covered during the training program. These meetings last approximately four hours and are held as many times as the company deems necessary. Although, obviously, there is no time to fully solve all the problems raised by the company engineers, the problems are sufficiently structured to allow the BP engineers to continue working on a detailed solution, based on the clear guidelines set by the consultants. When the problems are too complex to structure and solve during the work meetings, they are taken to the subsequent phase of the program, as described below.
Phase 4 is a consulting and follow-up activity addressed to various groups of engineers, depending on the area and the nature of the problem. During this phase, the University consultants offer the BP engineers technical and methodological support to structure and solve decision-making problems identified by the company. This type of advice can range between supporting them in completely structuring a specific decision-making problem, and the development, by the consultants, of a model to solve one of the company’s specific decision-making problems, in which case, the full solution is delivered to the company. In any case, the philosophy of the consulting activity is aimed at allowing the BP engineers to appropriate the various methodologies and models so that they can have the tools to solve by themselves decision-making problems with varying degrees of complexity.
Impact of the Training/Consulting Program
The training/consulting program described in the above section has had a significant impact in various BP areas such as Reservoir, Base Management, Budgets and Production. Figure 2 shows the training and consulting activities developed at BP and the company areas impacted by each of such activities.
Figure 2: Global Impact of the Training/Consulting Activities on BP
4. DESCRIPTION OF SOME PROBLEMS FOR WHICH CONSULTING HAS BEEN PROVIDED
This section offers a brief description of some of the problems borne by BP Colombia, which have been structured and developed using Decision Analysis methodologies, models and tools. The following problems are of various natures and varying degrees of complexity, thus requiring different models, a different way of working on them, and varying degrees of solution development. The complexity level varies because some of them are quite specific, as the one described in section 4.1, while others, such as the one presented under section 4.5, are more complex because they involve several company areas and may affect the company’s production strategy as a whole. The difference in the models used will be clearly seen in the description of each one of the problems we present in this section. The work methodology also changes because some of them are developed through a permanent work team with the BP engineers, as in the case of the problem described under 4.6, while other projects are totally developed by the University consultants based on specific requirements formulated by BP, as in the case of the problem described under 4.3.
In spite of the above, the solutions to all the problems described in this section have common characteristics: they have been subject to structured decision analysis methodologies, they have been analyzed with a probabilistic approach, and they have always been worked jointly with the BP engineers, validating the results as they are produced, throughout the various phases of each project.
4.1 Methodology and Models to Estimate and Analyze the Reserves of an Oil Well
Description of the Problem
BP Colombia did not have a technically supported methodology to obtain probability distributions and interest profiles regarding its oil production. Something similar happened with the estimation of the reserves of a specific well over a given period of time, based on identified relevant variables. The company wanted to run a risk analysis for annual production and reserves, obtaining the profiles for the 10%, 50% and 90% (P10, P50 & P90) percentiles, based on the most adequate probability distribution. In this case, we used a lognormal distribution, following the recommendations contained in the literature (See Capen  and Smith ).
Methodology and Models used to find a Solution
In order to solve the above-described problem, we designed a methodology with a probabilistic approach using the results of a specialized simulator (called VIP) to simulate the production of a well based on some variables related to the reservoir structure, such as: KH, skin and water volume, among others. Additionally, we applied the parametric method to obtain the probability distribution and the interest profiles of the production. The methodology encompasses the following steps:
1. Identify the relevant variables and the time horizon: We identified the relevant variables to estimate the production profiles and the well reserves, as well as the time period for which the analysis is done.
2. Define the base case: At this point, we estimated the most probable alternative for each of the variables, estimating the production profile and the reserves for the base case with the help of the VIP simulator.
3. Define the variable cases: We defined the alternatives to the base case for each variable and estimated the probability of each alternative.
4. Estimate the production profiles and the reserves: With the help of the VIP simulator, we estimated the production profiles and the corresponding reserves, pegging each of the variables to the various alternatives, and leaving the remaining ones in the base case.
5. Analyze the simulator results: Considering that the simulator uses the historical data of the different variables to produce results, we had to analyze the consistency of the estimated reserve profiles according to the experts’ opinion.
6. Analysis statistical processing: At this point we used the Parametric Method to obtain the probability distribution and the profiles corresponding to production P10, P50 and P90 and the reserves for every year. Furthermore, we analyzed the importance of each variable in estimating the production and the reserves. The above was done as follows:
Based on the simulation results, we established the expected value and the standard deviation of the impact of each of the relevant variables (on a percentage scale).
Assuming independence among the effects of each variable in the estimation of the reserves, we established the estimated value and the standard deviation of the global impact on the production and the reserves by multiplying the expected impacts of each variable.
With the global impact and standard deviation, we calculated the mean and the standard deviation of each year’s production in the original units.
Assuming that the reserves and the annual production behave according to a Lognormal distribution, we estimated the probability distribution and the P50, P10 & P90 annual production profiles and the reserves.
Finally, we established the importance of each relevant variable in the estimation of the production profiles and the reserves.
The developed methodology allowed us to obtain the probability distribution and the interest profiles of the oil production and of the reserves of a specific well for a given period of time. Such methodology is currently updated by the BP engineers.
4.2 Methodology and models used to estimate the IOR (Incremental Oil Rate)
Description of the Problem
BP did not have a technically supported methodology to analyze the behavior of the IOR in the short term (less than a year) of a well when some kind of wellwork is done, such as fracturing, re-perforating, and chemical stimulation, among others. Therefore, the goal was to design a methodology to estimate the most appropriate probability distribution to model the behavior of the IOR based on the behavior of some relevant variables related to the conditions of the reservoir, the conditions of the pipe and the surface conditions.
Methodology and Models used to find a Solution
In order to solve the above-described problem, we designed a methodology with a probabilistic approach using the results of a specialized software called Wellflow, to simulate the IOR of a given well, as well as the Montecarlo Simulation to obtain the probability distribution of the IOR. This methodology encompasses the following steps:
1. Identify the relevant variables: We identified the relevant variables to estimate the well’s IOR (KH, skin, and water volume, among others).
2. Define the base case: Here we estimated the most probable alternative for each variable, and this constituted the base case.
3. Define the range and the scenarios of the relevant variables: Here we defined the possible scenarios for each variable, in order to run a sensitivity analysis. In order to define the set of values to be allocated to each variable, we combined the experts’ opinions and the historical data. This allowed us to establish a probability distribution for the relevant variables and then generate the possible scenarios of such variables, based on such distribution.
4. Generate the IOR values: Once we defined the set of values for each relevant variable, we combined them to estimate the possible well IOR values, with the help of the Wellflow software.
5. Estimate the probability distribution of the IOR: With the values generated for the IOR, we used Crystal Ball to estimate the probability distribution that best fit the data. Insofar as possible, it is advisable to use a lognormal distribution, considering the recommendations of the papers that demonstrate that reserves behave according to this distribution (See Capen  and Smith ).
6. Obtain the annual IOR: Once we obtained the probability distribution of the IOR for the first month, we estimated the probability distribution for the monthly Declination Rate, in order to obtain the IOR for the following 12 months. We then obtained an average IOR for the year and, based on the annual declination rate, we estimated the IOR for the following year, plus any additional reserves.
7. Estimate the cost: Finally, we estimated the probability distribution for the cost of the work under evaluation and, using a spreadsheet, we estimated the relevant economic indicators such as net present value and capital efficiency. In order to estimate the cost of the well work, we designed the following probabilistic methodology:
The developed methodology allowed us to obtain the probability distribution for both the first month IOR and the average yearly IOR. Such methodology, as the one described under section 4.1, is currently being used by BP engineers in Colombia.
4.3 Methodology to Estimate the Cost of a Wellwork
Description of the Problem
When analyzing different types of wellworks such as fracturing, re-perforating and chemical stimulation, besides the probabilistic analysis of the IOR, one needs to estimate the cost associated to each specific wellwork, in order to rank the works according to their cost efficiency (cost/IOR).
Currently, BP performs an average of between 170 and 180 wellworks per year, at an estimated annual cost of US $19 million. The cost of each wellwork depends on the activities or tasks performed, and the duration of each activity. Additionally, there are indirect costs such as salaries and the use of each piece of equipment, which must be distributed proportionally among all the wellworks.
The project consisted in designing a methodology with a probabilistic approach to estimate and analyze how the cost behaves for each type of wellwork.
Methodology and Models Used to Find a Solution
In order to solve the above-described problem, we designed a methodology based on a work plan designed by BP for each month, which takes into account both the direct cost of each wellwork and the distribution of the costs shared by several wellworks. This information was used to estimate the total cost of each wellwork. The following are the steps comprised by the methodology:
1. Identify and clarify the monthly work plan prepared by BP.
2. Identify the type of work to be evaluated.
3. Identify the activities or tasks corresponding to such works.
4. Determine the possible results that might be obtained when performing each of the activities identified in the prior step, and estimate the probability of each such result, based on historical data and on the experts’ opinion. This information is then plotted on a graph illustrating the sequence of all such activities.
5. Estimate the probability distributions for the time required by each activity. Here you use triangular distributions and estimate their parameters based on the historical data and the experts’ opinion.
6. Establish the unit rates (per hour, per day, per barrel, etc.) to be used when estimating the costs.
7. Estimate the costs of the various activities or groups of activities, based on the unit rates and the time required by each activity.
8. By using the Montecarlo Simulation and the tool described below, estimate the probability distribution of the total work time and the probability distribution of the total work cost.
Besides the methodology, we designed the WellWorks Risk tool, which runs on Microsoft Excel. We used this tool and the Crystal Ball simulation software to estimate and analyze the behavior of the wellwork cost after following the steps of the methodology presented in the above section. The main menu of this tool is shown on Figure 3.
Figure 3: Main Menu of the WellWorks Risk Tool
The methodology and the tool we developed, which are already being used by BP, allow for a probabilistic analysis of the time and cost required for a given wellwork, the probability distribution, the basic statistics and the percentiles associated to such variables. By way of illustration, Figure 4 shows a very simplified example of a wellwork with a specific sequence and the type of results produced by the methodology. It is worth mentioning that in the actual cases dealt with by BP, a typical wellwork may have between 40 and 80 activities.
Figure 4: Results of the WellWorks Risk Tool for a Specific Example
4.4 Decision-Making Model for the Exploitation of an Oil Well
Description of the Problem
BP was interested in defining the best strategy to drill and exploit a new well in one of its fields, for which it had to evaluate the various drilling and injection alternatives and compare them with some kind of economic performance parameter.
Methodology and Models used to find a Solution
In order to solve the above-described problem, we structured a decision-making tree, identifying the decision variables, the random variables and the relevant economic information required to evaluate the various alternatives.
1. Decision-making variables:
- Injection well: There were 3 alternatives for gas injector wells to increase oil production. This decision conditions the drilling strategy to be applied to drill the producer well.
- Formation to reach by injection: For some of the injector wells, the company must decide the formation in which it will inject the gas; if it injects it into Mirador, which bears the greatest success probability, or into Barco, in which case the success probability would be less, but the oil production would grow.
- Drilling site: Once the injection strategy is defined, BP must decide whether it drills the producer well on the flank, either from an existing pad or from a new pad, or on the top of the formation.
- Formation to reach by drilling: If the well is drilled on the flank, only the Mirador formation can be reached, while if it is drilled on the top of the formation, the company has the choice of drilling only until Mirador or of continuing on to reach Barco, which would enhance the production rate.
2. Random variables:
The results of the decisions described above are random Bernoulli variables with two possible results: success or failure. The other random variable is the reserves level obtained under the two different decision alternatives. This variable was modeled as a discreet random variable with three possible levels: high, medium and low. The probability functions associated to the various random variables were estimated by the BP experts based on their specific knowledge of the problem and on historic information.
3. Relevant economic information:
The performance parameter used by BP to evaluate its projects is cost efficiency (cost/oil). Nevertheless, in the case of this problem it was not possible to use this performance parameter, because in some cases the reserves obtained may be equal to zero. So we used instead the parameter of net profit (reserves x price – costs), supposing a price of US $20 per barrel. The reserves obtained under the various alternatives, as well as the costs associated to each reserve level, were estimated by company experts based on their experience and with the help of the VIP simulator.
Once we defined the decision variables, the random variables, the relevant economic information and the sequence of the decision-making process, we structured the decision tree shown in Figure 5. Because the tree is so big, it was not possible to show it in a single graph; thus, the tree shown on the left, containing the terminal nodes with the annotation “Perforation Tree” would be attached to the right hand tree, changing the economic values and the probabilities, according to the case.
Figure 5: Decision-Making Tree for the Exploitation of an Oil Well
BP experts are currently estimating the costs and the reserves associated to each of the alternatives, in order to evaluate the tree and obtain the optimum policy and the risk distribution associated to such policy.
4.5 Methodology to determine Gas Production Profiles
Description of the Problem
Faced with the opportunity to produce fuel from gas by using a new technology, BP needed to estimate the amount of gas it could produce in its three fields in Colombia during the coming years. The company set a target production level and wanted to find out the year until which it could meet such production level. Considering some commercial commitments, fields 1 & 2 could begin producing gas immediately to convert it into fuel, and field 3 would begin producing gas when the production of the first two became insufficient to meet the target.
Methodology and Models used to find a Solution
To solve the above-described problem, we designed a probabilistic methodology to find out the probability distributions of the gas production during the coming years and of the oil that would not be produced when producing the gas. Likewise, the methodology allowed us to estimate the time during which the production target set by the company could be met. This methodology, based on models such as the Parametric Method and the Montecarlo Simulation, encompasses the following steps:
1. Run the simulations for fields 1 and 2 to obtain the probability density function (pdf) of gas production for each year using the Parametric Method.
2. Add up fields 1 y 2 gas annual production using Montecarlo Simulation and approach these annual distributions with discrete probability functions using the Three Points Approximation . This is a method that allows approaching a probability density function of a continuous random variable with a probability function of a discrete random variable.
3. Run the simulations for field 3 to calculate the pdf of gas annual production, starting from each of the three profiles obtained in Step 2.
4. Add up the annual production of the three fields to obtain the pdf of total gas annual production using Montecarlo Simulation.
5. Using a specific methodology that use the Gas Oil Rate (GOR), estimate the total Black Oil Lost (BOL) as a function of the total gas annual production.
6. Obtain final results:
i) The pdf of the total gas annual production.
ii) The P90, P50 and P10 profiles of the total gas annual production.
iii) The pdf of the Demand Satisfaction Period (DSP) with the respective percentiles.
iv) The BOL for the profiles of interest of the total gas annual production.
When the developed methodology was applied, we obtained the results described under step 6 of the methodology. Figure 6 shows the profiles for production at P90, P50 and P10 during each year, as well as the 10th, 50th and 90th percentiles of the DSP for specific data .
Figure 6: Results of the Methodology to determine Gas Production Profiles
4.6 Gas Production Strategies
Description and Structuring of the of the Problem
BP is evaluating the possibility of defining a production strategy to increase gas production in two of its youngest fields. Considering the complexity of the problem, given the number of alternatives and the variety of the decision-making criteria, we structured the problem in detail, identifying its main aspects, the decision-making criteria and the initial alternatives for a solution.
- Technical Aspect: This refers to the various production technologies.
- Economic Aspect: This refers to the costs and various investment levels of the different production alternatives, as well as their variations, which depend on whether BP’s partners participate or not.
- Commercial Aspect: This includes the rates to be paid to the partners under the various contracts for some of the production alternatives.
- Strategic Aspect: This refers mainly to the advisability of the various production alternatives regarding other BP projects, and the strengthening of the company’s operations in Colombia in the medium and long terms.
- Economic: NPV, Time required to recover the investment, etc.
- Reserves recovered.
- Synergies with other projects of importance for the company.
- Sustainability of BP’s business in Colombia.
Initial Decision-Making Alternatives
For field 1, BP has initially identified the following alternatives:
- Increase the injection level, drill new injector wells (number and site to be determined) and increase compression to inject. This implies enlarging the plant, at an associated cost.
- Interconnecting fields 1 & 2, at a given cost. This alternative implies a decision as to the amount of gas to be injected.
- Do nothing.
For field 2, BP has initially identified the following alternatives:
- Enlarge this field’s plant to raise its gas production capacity, at a given cost.
- Considering that enlarging the current plant is not enough, the alternative is to build an additional plant, besides enlarging the existing one. Nevertheless, this alternative raises the backpressure, which would affect oil production.
- Do nothing.
BP is currently analyzing the various alternatives and comparing this decision-making problem with other BP projects, in order to proceed to choose a solution.
5. RESULTS AND CONCLUSIONS
The most important conclusion regarding the above-described experience is the global impact that the projects developed have had on BP – Colombia’s decision-making culture and, specifically, on the procedures used to analyze and make decisions in the Reservoir, Base Management, Budgets and Production areas. The authors know of no similar experience in this field in Colombia, except for the project that this same consulting group has been developing at ECOPETROL, the Colombian state-oil-company. For the University consultants of the Los Andes University, it has been a demanding but encouraging experience, since it has allowed them to identify especially relevant applications of the Decision Analysis methodologies, tools and models applicable to actual situations of a company operating under international standards. Furthermore, there are clear indications that some developments achieved by BP – Colombia in this area would be incorporated into BP International’s decision-making and risk-analysis practices.
The organization has learned that by having a methodology to understand uncertainties and rigorously assess risk, the quality of decisions made has improved dramatically. Today, an important group of engineers understands much better several models and methods that they used and applied mechanically before the training program, without understanding how they really worked.
The fact that risk & uncertainty are being managed differently now has given way to different conversations between project owners and decision makers. Said conversations are much more focused and hence decisions are much more informed. As a result, projects are now being better prioritized for resource allocation purposes.
Innovation came with the inclusion of (academic) tools in the standard toolkit used for decision-making. The combined effect of new tools and real hands-on experience has resulted in innovative ways to manage and communicate operational and business risks to decision makers.
 Capen E.C. (2001), “Probabilistic Reserves! Here at Last?”, Hydrocarbon Economics and Evaluation Symposium, Dallas, 1999.
 Castillo, M. (2004), “Toma de Decisiones en las Empresas: Entre el Arte y la Técnica” (to be published soon).
 Castillo, M., y A. J. Mendoza , “Metodología de Decisión para determinar la Mejor Estrategia de Operación de un Sistema de Poliductos en ECOPETROL”, Working Paper.
 Castillo, M. y G. Duque y F. Beltrán (2002), “Incentives: Mechanisms for the Consistency of the Colombian Educational System”, Kybernetes, Vol. 31, No. 9/10, pp.1282 – 1288.
 Castillo, M. y M. Jiménez (2004), “Metodología para determinar la Viabilidad Técnica y Financiera de un tercer Nodo para las Operaciones del Banco de la República”, Revista de Ingeniería No. 19, Universidad de los Andes.
 Castillo, M. y A. J. Mendoza (2004), “Diseño de una Metodología para la Identificación y la Medición del Riesgo Operativo en Instituciones Financieras”, Revista de Ingeniería No. 19, Universidad de los Andes.
 Clemen, R.T. (1996), “Making Hard Decisions: An Introduction to Decision Analysis”, Duxbury Press.
 Evans, J. y D. Olson (1998), “Introduction to Simulation and Risk Analysis”, Prentice Hall.
 Fetkovich, M.J., E.J Fetkovich y M.D. Fetkovich (1996), “Useful Concepts for Decline-Curve Forecasting, Reserve Estimation, and Analysis”, SPE Reservoir Engineering.
 Keefer D.L. y S.E. Bodily (1983), “Three Point Approximations for Continuous Random Variables”, Management Science, Vol. 29, No. 5, p.p. 595-609.
 Matheson, D. y J. E. Matheson (1998), “The Smart Organization: Creating Value Through Strategic R&D”, Harvard Business School Press.
 Nadkarni, S. y P. Shenoy (2000), “A Causal Mapping Approach to Constructing Bayesian Networks”, School of Business Working Paper No. 289, University of Kansas.
 Neapolitan, R. E. (2004), “Learning Bayesian Networks”, Pearson Prentice Hall.
 Rose, P. (1987), “Dealing with Risk and Uncertainty in Exploration: How Can We Improve?”, The American Association of Petroleum Geologists Bulletin, Vol. 71, No. 1, p.p. 1-16.
 Saaty, T. (1994), “Fundamental of Decision Making and Priority Theory with the Analytic Hierarchy Process”, University of Pittsburgh.
 Smith, P.J., D. J. Hendry y A. R. Crowther (1993), “The Quantification and Management of Uncertainty in Reserves”, SPE 26056, Western Regional Meeting, Anchorage, USA.
 von Winterfeldt, D. y W. Edwards (1986), “Decision Analysis and Behavioral Research”, Cambridge University Press.