Análise comparativa entre métodos de reconhecimento de padrões aplicados ao problema Gas-Lift intermitente

by Júnior, Oswaldo Ludwig

The purpose of this work is to compare statistical techniques and connectionist ap-proaches in the recognition of fail patterns on an oil extraction process that receivesthe "Gas-lift"denomination. Connectionist approach is based on feedforward mul-tilayer artificial neural network and self-organizing maps. The obtained results arejustified by the VC dimension analysis. Alternative methods to features extractionproblems are also applied. This work contributes to the current state of the art by theoptimization of the Kohonen networks through the use of the Battacharyya distance.It is also contribution of this work the use of the entropy level in the characteris-tics extraction process. The obtained results are susceptible to comparison with theresults from the Project SGPA II, developed by the Department of Mechanical Engi-neering of UFBA, with the purpose of to complement and to validate the two studiesmutually.