Method for producing an optimized neural network module for simulating the flow mode of a multiphased vein of fluids

摘要

Uses hydrodynamic and/or thermodynamic magnitudes. The module is integrated into a general module for both thermodynamic and hydrodynamic multiphase liquid flow simulation. The model is used to form a learning base so as to select from the physical magnitudes the best for model operation, as well as the variation range fixed for parameters and magnitudes. The network results adjust themselves to the best formed learning base. Method of forming a module designed to simulate in real time the flow mode at any point in a conduit for multiphase fluid flow including a liquid and gaseous phase. The method optimizes for the best fixed operating conditions constrained by structural parameters relative to the conduit using a group of defined physical magnitudes with fixed variation range for the parameters and magnitudes. The modeling system includes a non-linear neural network base with each of the inputs for structure parameters and physical magnitudes and with outputs giving the necessary magnitudes for the estimation of the flow mode. The network includes at least intermediate layer. The network is determined iteratively so as to adjust the values of the learning base with pre-defined tables linking different values produced for the data output to values corresponding to input data. The learning base is designed for imposed operating conditions and the neural network is generated to adjust itself to the best imposed operating conditions.