Third International Conference on Multiphase Flow, ICMF98
Lyon, France, June 8-12, 1998



COOPERATION BETWEEN FLUIDYNAMIC AND NEURAL-FUZZY MODELS FOR MONITORING MULTIPHASE FLOWS

Annunziato M. , Bertini I. , Pizzuti S.
E.N.E.A. - Via Anguillarese 301, 00060 S.Maria di Galeria (Rome), Italy


Abstract

In most of industrial applications and in the fields of scientific research often phenomena for which it is hard to formulate correct models are studied. The main reason for this is that the underlying phenomena are highly non-linear and/or they have high dimensionality. In such cases a model which describes exactly the phenomenon does not exist, but often many partial models describing the problem's phenomenology in particular conditions are available. At present the most utilised approach to solve such problems is that of comparing all the available models and techniques and then choose the one which behaves better than the others in all different conditions. The problem of the multiphase flow rate estimation in oil extraction and transport processes fills in with this class of problems. In fact the approach used at present in this field is essentially that of studying new physical models and new advanced techniques and to choose the best one.
In our paper we propose an innovative approach for multiphase flow rate estimation. Our approach is inspired by an innovative principle of theoretical physics known as Bootstrap Theory which proposes the integration of all different models building up a cooperative system able to get the best features of the models in all different conditions. The task of such approach is that of getting a system which performs better than the best available model. It is clear that the attention of research moves from the study of the models to the relationship among the models. In this paper we describe how we applied the 'Cooperative Bootstrap Approach' to the problem of the estimation of multiphase flow rates. In our work we built up a system in which mathematical models of multiphase flow rate estimation cooperate with neural networks by using a meta-decision maker based on fuzzy set theory. The developed system has been successfully installed in the AGIP oil field (Trecate - Italy) and it is inserted within the C.E. Thermie project OG/0143/94 "Monitoring and diagnostic system, based on expert system technology, for multiphase transportation processes" , leader : ENEA, partners: AGIP, Rome 1st University, Gammatom.
The innovative aspects of the work are the theoretical study of how different techniques (fluidynamic models and neural networks) can cooperate and the application of such complex systems to the estimation of multiphase flow rates.
The reason to let neural networks and fluidynamic models cooperate, is that of trying to get the best features from both. In particular neural networks are extremely precise in their training set because they are able to get the highly non-linear aspects of the phenomenon. The drawback of this technique is that neural networks are black box so their results cannot be scaled to different conditions. On the contrary the fluidynamic models do not get precisely all the non linear behaviours but they can be scaled to different operating conditions.
The results of all the modules provide the decision module. This module, or data fusion module, is a fuzzy logic based system which provides the final flow rates estimation. Its core is the suitable definition of measure reliability based on opportune fuzzy rules. The criteria, on which these rules are based, are the measure error estimation and the measure neighbourhood to the neural network training set. The final choice will favour more the most reliable measure and less the other ones. Moreover the problems of the choice of the opportune membership functions and that of error estimation have been carried out.
Various architectures of such system have been tested on real data and the best results provide very low final errors for the flow rates estimation showing a remarkable improvement in such estimations and thus a considerable error decrease. In particular the system has been tested on two bad fluidynamic models for liquid and gas estimation. The system has cut down the measurement error from 12% to 4%, in the case of liquid estimation, and from 20% to 4% in the case of gas estimation.
The cooperative approach has also allowed us to build a scaleable system. In fact with the opportune fuzzy rules it is possible to let the system work in condition far from the training ones and in conditions of sensors’ failures. Finally each module can be viewed as a virtual sensor so the proposed architecture can be generalised to an arbitrary number of different measurement systems making it possible the simulation and the experimentation of new sensors without having them physically.