In the present study, the neural network approach is applied to predict the flow rate of a three-phase mixture of oil-water and gas through a Venturi meter. Different types of nets have been compared searching for an optimal configuration. The aim is make a comparison with the physical models previously developed trying to improve the prediction accuracy. The models and the neural network have been tested on a data set obtained on a experimental plant with real processing fluids. An different scheme of pressure difference measurements on the Venturi tube has been used, trying to obtain more information. The chosen scheme and the supplied information on fluid densities seems to be sufficient for the NN to predict correctly the flow rate of each phase. The standard deviation of relative errors are less than 5% and the average error is very close to zero. In previous studies the physical models are able to predict quite correctly the flow rates (average error ~0%) but the accuracy was worst (~15%).
USING NEURAL NETWORK IN MULTIPHASE METERING / Alimonti, Claudio; Bilardo, Ugo. - ELETTRONICO. - (2001). (Intervento presentato al convegno 4th Int. Conf. on Multiphase Flow tenutosi a New Orleans LA nel May 27-June 1).
USING NEURAL NETWORK IN MULTIPHASE METERING
ALIMONTI, Claudio;BILARDO, Ugo
2001
Abstract
In the present study, the neural network approach is applied to predict the flow rate of a three-phase mixture of oil-water and gas through a Venturi meter. Different types of nets have been compared searching for an optimal configuration. The aim is make a comparison with the physical models previously developed trying to improve the prediction accuracy. The models and the neural network have been tested on a data set obtained on a experimental plant with real processing fluids. An different scheme of pressure difference measurements on the Venturi tube has been used, trying to obtain more information. The chosen scheme and the supplied information on fluid densities seems to be sufficient for the NN to predict correctly the flow rate of each phase. The standard deviation of relative errors are less than 5% and the average error is very close to zero. In previous studies the physical models are able to predict quite correctly the flow rates (average error ~0%) but the accuracy was worst (~15%).I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
