This paper represents an integration of artificial intelligence and lean sigma techniques to achieve large field production optimization.The first part of the methodology (detailed in SPE 90266 "Zonal Allocation and Increased Production Opportunities Using Data Mining in Kern River"[1]) involves data management and predictive data mining for increased production opportunity identification.It utilizes a set of data mining tools including clustering techniques and neural networks to identify new candidates for clean-outs, perforating, sidetracks, deepening, and other types of workovers.Furthermore, the expert system was used to predict the estimated production increase for these candidates.The second part of the methodology optimizes the implementation and post-workover follow up of the opportunities identified in part one.It involves the use of lean sigma tools such as value stream mapping, level loading, continuous flow production, standard operating procedures, and kanbans which optimize execution cycle time, peak oil production, decision making process, cost, and safety[2].This approach was successfully applied and executed in the Kern River field. Introduction With over 8,600 active producers averaging 10 BOPD each and a limited staff, streamlining the well optimization process in the Kern River field is critical to take advantage of a large and dynamic portfolio of relatively low oil gain opportunities.It is essential to effectively identify, prioritize, and implement a high number of these opportunities, which typically range from 2 to 8 incremental barrels of oil per day. As detailed in SPE paper 902661, a significant production increase opportunity was discovered in the lower sands through the use of artificial intelligence tools after observing that some wells in the field have high production, while nearby neighbor wells are very low producers.A pilot program was implemented and following its success, the study was extended across the entire field.After identifying the field-wide opportunity, a significant workover program was launched. A lookback on the pilot program indicated several processes, including candidate selection, were successful and would continue to be used "as is" in the execution of the field wide effort.The post-workover follow up and put on production (POP) processes, however, were identified as weaknesses and were highlighted as areas of improvement.Lean Sigma techniques were selected to optimize and streamline these processes. Background This paper represents an integration of artificial intelligence and lean sigma techniques to improve workflow processes and execution of a large field optimization project in Kern River. Reservoir Description.The Kern River field, located in Kern County, California, is a heavy oil reservoir consisting of nine productive sand intervals and many more individual sand lobes or flow units within the Kern River series.The field is 4 miles by 5 miles in areal extent and has over 8,600 active producing wells and 1,200 steam injectors.Producers are co-mingled with very little individual zone production test data available.The field is currently produced by steam injection with varying degrees of thermal maturity in each of the sands.The primary production mechanism is gravity drainage with extremely low average reservoir pressure of 20 psi in the oil sands, requiring pumps to be set at or below the bottom-most oil sand and pumped off to effectively produce. The northeast half of the field has little to no water impacted sands, while the lowermost sands in the central portion of the field are water/aquifer impacted.The water impacted sands are found progressively higher moving southwest, down structure, across the southwest half of the field.Higher pressures exceeding 50 psi are found in these sands.
Case-based reasoning, also known as computer reasoning by analogy, is a simple and practical technique that solves new problems by comparing them to ones that have already been solved in the past, thus saving time and money. The technique constantly incorporates dynamic data, which empowers the system to learn and adapt from new experiences. A general framework for case-based reasoning is presented, along with a review of the four-step cycle that characterizes the technology: retrieve, reuse, revise and retrain.This paper presents a specific application of case-based reasoning to determine the optimum cleaning technique (bailing, washback, or foaming) for sanded/seized well failures. The methodology extracts only the most relevant information from the historical database, utilizes a rule-based system to make adaptations, and then suggests the most appropriate solution for a given well intervention. This technique was used for production operations as a front end tool for well workover planning and design and was applied to sample data from a large oil field where the main artificial lift system is rod pump. This simple case demonstrates how case-based reasoning can be applied to improve the planning and execution of well interventions, thus reducing cost, rig time, and well turnaround time, while maximizing reliability and production.
One of the simplest ways to increase production from a well is to optimize the pumping unit. A properly sized and configured pumping unit lifts the maximum barrels of fluid that the well is capable of producing while minimizing the wear on the rods and pump. Optimization is usually done through the use of software such as RODSTAR or ECHOMETER on a well-by-well basis. The user must input all the well parameters by hand, causing it to be a long and tedious process. This makes overall field-wide optimization extremely time-consuming for fields in which there are many wells. Moreover, the software determines theoretically the most optimal unit configuration for the well and may recommend a unit that is not readily available in the area. This paper describes a new approach for field wide pumping unit optimization. The methodology described here uses a neuro-system consisting of a neural network and an intelligent swapping procedure to find the optimum pumping unit placement for the field. Since field data is used in the model, only pumping units readily available in the field are used. This method is suitable for any field with pumping wells, especially those fields in which there are many wells, thus allowing the selection of a confident data set. Criteria and constraints are set to select the wells that are currently sufficiently optimized. These wells are used as the model data to train a confident neural network. Optimum pumping unit sizes are then predicted for those wells that are considered to be non-optimized. Finally, an intelligent swapping procedure is invoked to swap over- and under-sized units, thus providing field wide optimization. The final goal of this study is to provide a tool that allows engineers to set acceptable, realistic criteria to optimize pumping unit size for each well and pumping unit placement on a field-wide basis. The paper presents an example of methodology applicability to a Chevron-operated oil field in California. The proposed procedure provides confident results, great flexibility, and fast optimization. Introduction Rod-pumped production is one of the oldest and most practical methods of producing reservoirs. It is still wildly used today in many fields. However, to produce the maximum amount of fluid, the pumping unit must be optimized for the potential of the well. Pumping unit optimization has become a necessity for the industry. As a result computers programs such as RODSTAR and ECHOMETER are used to help in the selection of the best unit to fit the production potential of the well. This is oftentimes not a very practical approach since the suggested unit may not be available in the field. As a result, a similarly sized unit will be placed on the well. To compound the problem, the potential of a well is always changing over time due to factors such as depletion, formation damage, stimulations, and workovers. Even if the pumping unit has been optimally sized at initial completion, the well may likely become non-optimized over time as the potential of the well changes. The problem addressed in this work is field wide pumping unit optimization using only the available units in the field. In other words, finding an optimum placement combination of the units existing in the field for currently non-optimized wells. Usually pumping unit optimization is done one well at a time. This well-by-well optimization is time consuming and requires a certain amount of specialization for this kind of task. Moreover, a human being can only perform a simple swapping procedure. Usually this is no more than the simple exchange of units between two wells. Complex swapping scenarios to optimize the entire field are not possible by these means.
TX 75083-3836, U.S.A., fax 01-972-952-9435. AbstractOne of the simplest ways to increase production from a well is to optimize the pumping unit. A properly sized and configured pumping unit lifts the maximum barrels of fluid that the well is capable of producing while minimizing the wear on the rods and pump. Optimization is usually done through the use of software such as RODSTAR or ECHOMETER on a wellby-well basis. The user must input all the well parameters by hand, causing it to be a long and tedious process. This makes overall field-wide optimization extremely time-consuming for fields in which there are many wells. Moreover, the software determines theoretically the most optimal unit configuration for the well and may recommend a unit that is not readily available in the area.This paper describes a new approach for field wide pumping unit optimization. The methodology described here uses a neuro-system consisting of a neural network and an intelligent swapping procedure to find the optimum pumping unit placement for the field. Since field data is used in the model, only pumping units readily available in the field are used. This method is suitable for any field with pumping wells, especially those fields in which there are many wells, thus allowing the selection of a confident data set.Criteria and constraints are set to select the wells that are currently sufficiently optimized. These wells are used as the model data to train a confident neural network. Optimum pumping unit sizes are then predicted for those wells that are considered to be non-optimized. Finally, an intelligent swapping procedure is invoked to swap over-and under-sized units, thus providing field wide optimization.The final goal of this study is to provide a tool that allows engineers to set acceptable, realistic criteria to optimize pumping unit size for each well and pumping unit placement on a field-wide basis. The paper presents an example of methodology applicability to a Chevron-operated oil field in California. The proposed procedure provides confident results, great flexibility, and fast optimization.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.