Abstract: The well test is one of the primary diagnostic andpredictive tools used in the analysis of oil and gaswells. In these tests, a pressure recording device isplaced in the well and the pressure response isrecorded over time under controlled flow conditions.The interpreted results are indicators of the well'sability to flow and the damage done to the formationsurrounding the wellbore during drilling andcompletion. The results are used for many purposes,including reservoir modeling (simulation) and economicforecasting. The first step in the analysis is theidentification of the Well-Test Interpretation (WTI)model, which determines the appropriate solutionmethod. Mis-identification of the WTI model occurs dueto noise and non-ideal reservoir conditions. Previousstudies have shown that a feed-forward neural networkusing the backpropagation algorithm can be used toidentify the WTI model. One of the drawbacks to thisapproach is, however, training time, which can run intodays of CPU time on personal computers. In this paper asimilar neural network is applied using both a personalcomputer and a neurocomputer. Input data processing,network design, and performance are discussed andcompared. The results show that the neurocomputergreatly eases the burden of training and allows thenetwork to outperform a similar network running on apersonal computer.!9
展开▼
