This paper will present a software that was developed to diagnose well test data. The software monitors the data, and through a series of algorithms alarms the user in case of discrepancies. This allows the user to investigate possible source of errors and correct them in real time. Several datasets from previous operations were analyzed and the basic physics governing how a certain datum depends on others were laid out. All the well test data traditionally acquired were put on a matrix, showing the dependencies between each datum and other physical properties that are available - either measured or modelled. Acceptable fluctuations in acquired data were also identified for use as tolerance limits. The software scans through the data as it is acquired and raises an alarm when the identified dependencies are broken. The software also identified which parameter is most likely causing the error. The software was built based on previous well test data and reports. Subsequently, two field trials were conducted to fine tune the algorithms and allowable data fluctuations. The process of validating the software consisted of: (1) Identifying flagged errors that should have not been flagged (dependencies set too tight); (2) identifying errors that should have been flagged and were not (dependencies set too loose); (3) improving the user interface for ease of use. The results were positive, with several improvements in the error recognition and several discrepancies flagged that would not have been caught by the naked eye. The user interface was also improved, allowing the user to clear error messages and provide input to improve the algorithm. The field trial also demonstrated that the methodology is scalable to other data acquisition plans and to more advanced analytics. The algorithms are simple, allowing the software to be implemented in all operations. More advanced algorithms are likely to depend on job specific data and parameters. Traditional data acquisition systems used during well test only present the data. Alarms trigger the user's attention only when certain defined operability limits are about to be reached. Being able to confirm that the data is cohesive during the well test prevents a loss of confidence in the results and painful post processing exercises. Moreover, given the algorithms used are based on simple physics, it is easy to deploy the software in any operation.

No abstract

No abstract

No abstract