The differences between foreseen results in lift well project related to the obtained in production phase increases preventing and correcting actions necessary to optimize well production. The larger number of wells and its analysis complexity don't allow to conduct the cluster of actions entirely. To help in this task, the automation of petroleum wells has grown expressively. Over 700 wells in Brazil, at Bahia State, are provided with PLCs using A.I. techniques for their control. This sort of automation generates an expressive amount of data that can be used to optimize the lift systems production. Besides this, it also allows local control and remote observation of many wells in real time basis. This paper presents an approach for an automated lift well management system, called SGPA. SGPA supports diagnosing and proposes solutions to optimize automated wells production. SGPA focus mainly on Rod-Pump wells management. However, other lift systems can be included as well. SGPA operates under three main subsystems:Data consistency module, makes the acquisition and consistency of information from different sources. This information is treated and shown in a friendly interface, and it is used in other subsystems.Dynamometric Cards apprenticeships module, allows the user to generate and train the system understanding of the dynamometric cards pattern behavior. These patterns can be used by the PLC's which control the well, and by the SGPA pattern recognition module to support the well diagnose and improve the well lift.Knowledge module that makes well analysis. The knowledge subsystem determine symptoms, makes diagnostics and adds intelligence for the well optimizing and correcting actions. The knowledge representation uses Symbolic Neural Networks (SNN) and Fuzzy Logic (FL) for its constructions and apprenticeship and evolution. It is understood that lift systems automation comes to be a complex system that distinguishes from other traditional approaches by successfully dealing with huge amount of information to cope its mission, despite the high level of empiricism and apparently erratic behavior due to well peculiarities, demanding rigorous status observation. Such reality and the present technology state-of-art justify SGPA development. Introduction Oil wells have two characteristic phases in it's production life, design and follow-up. In the design phase equipment are calculated and specified based in estimated data. In the follow-up phase estimated values are compared to measured values and design corrections should be made. In this last phase, design optimization should be implemented to reduce costs, increase production and raise the mean time between failures (MTBF). The equipment design based in poor data leads to an oversized equipment and to the necessity of redesigning the lifting system based in the measured data. This demands the analysis of a huge amount of well data, that become known as the time goes by. Usually the necessity of well servicing happens to happen in an unexpected day, where this analysis should be made in a few hours. The huge amount of data and analysis complexity causes the lift system redesign to be made only in part. Each lift system has different behavior due to different well characteristics what makes the use of generic solution a difficult task. These facts causes the lift system efficiency to be decreased by constants production stops without no easy reason identification.

No abstract

SGPA is an Intelligent Distributed Management System for Automated Wells developed back in 2001/2002 for sucker-rod pumped wells analysis and management. The SGPA concepts and ideas were presented in the 2002 SPE ATC&E, in San Antonio, Tx. SGPA operates under three main subsystems, Data consistency module for acquisition and consistency of the information from different sources, Knowledge module for well analysis based on Artificial Lift experts coded experience and dynamometer card apprenticeship module for dynamometer card patterns development. The sucker rod pumping experts knowledge were coded, used and improved in several well analysis creating a Sucker rod pumping knowledge base used now to do more precise analysis and also to teach and train new engineers. Well automation brought a big change to the field personnel routine, in a few months a huge amount of well information became available and the field personnel urged to use it to implement well analysis. The dynamometer card patterns to be used by the PLC at the well site for optimization and control was also an issue. The use of patterns for normal operation, pump-off, gas lock, etc… is a key feature of the well automation system and a fundamental necessity in any well analysis. The SGPA dynamometer card apprenticeship subsystem was used to develop dynamometer patterns based in the acquired real dynamometric cards stored in the well automation system databases. This paper presents an in depth description of the SGPA knowledge base after one year use and the set of dynamometric card patterns developed. Introduction SGPA is an intelligent distributed system with three distinct levels: Local, where the actions over the well are performed by a Programable Logic Computer (PLC); Supervision, where human actions, restricted to emergency actions are performed; Analysis, where human actions deals with the application of the diagnostics and therapies proposed by the SGPA intelligent module. [3]

The Progressing Cavity Pump artificial lift method for oil well production is well known in the oil industry, despite that a poor literature about PCP analysis is available. PCP automation based on ladder programming deploys information such as instantaneous rod torque, production line pressure, and rods rotation to name a few. The use of this information to analyze the well performance, identify imperfections and problems depends on a deep knowledge about the system well-PCP, but the procedures to do so are yet to be developed. To optimize the production and increase the mean time between failures in this type of well is necessary not only to improve the algorithm coded on the ladder programming loaded in the PCP automation PLC but also to develop a knowledge base where the skills and knowledge of the local PCP specialists could be easily coded. To analyze data deployed by the PLC and to improve the firmware customization, an intelligent system is being developed. This intelligent system is based on artificial intelligence concepts such fuzzy functions, coded in a way to easy acquire the PCP specialist knowledge on how to understand what is going on in the well and what to do to bring it back to a proper behavior. The present paper presents an intelligent system for automation of PCP wells capable of acquire and spread the knowledge of the referring specialist in diagnosis and solutions of problems during the well operation. Fuzzy agents are proposed in order to do an intelligent analysis of a set of PCP variables such as friction, motor, pump position, production and volumetric efficiency, ... The paper also describes the tools and interfaces designed to acquire the PCP specialist knowledge, the system implementation, and an acquired knowledge base example. Introduction Progressing Cavity Pumping (PCP) control is simpler than the control used in Sucker Rod Pumping (SRP) where a customized firmware is needed and demands a far more complicated analysis [4,6]. All the functionality needed to implement the PCP control can be obtained using ladder logic programming language. Ladder logic programming language is a generic programming language to be used in Programmable Logic Controllers (PLC) and is a well know and low cost development programming language. All the well data generated by the Ladder Program, see figure below, are collected by a communication driver and stored in a Data-base at the central office, in a non stop sequential pooling. This data-base is merged to others data-base such as production and well servicing data-bases through a corporate network. The resulted data-base is the data-source to the analysis presented in this paper. A Mamdani fuzzy structure [1] was chosen to represent the knowledge needed to analyze all the data gathered. The Artificial Lift knowledge collected from Artificial Lift Specialists (ALS) was broken in to small pieces and coded as Fuzzy Agents (FA), e.g. specialized agents to analyze hypothesis about well characteristics that has impact on the PCP performance such as friction, gas interference, motor HP, pumping efficiency, pump depth, paraffin, scale, sand deposition and corrosion [6]. This work extends the PCP analysis presented by Carvalho et alli [5] to an all Fuzzy approach. Fuzzy Concepts There are two popular fuzzy structures commonly used: the Mamdani [1] and Takagi-Sugeno (TS) [2] fuzzy structure. In this paper, the first one is adopted.