MODULAR NEURAL NETWORK MODELING FOR REFRACTIVE INDICES OF SOME BINARY SYSTEMS
GABRIELA LISA *, CĂTĂLIN LISA, SILVIA CURTEANU „GH. ASACHI” TECHNICAL UNIVERSITY IAŞI, FACULTY OF CHEMICAL ENGINEERING AND ENVIRONMENTAL PROTECTION, B-DUL D. MANGERON, NO. 71, 700050 - IAŞI, ROMANIA
In this work, a modular neural network (MNN) model for prediction of the thermodynamic properties was established. The numbers of patterns used in this study were 175, 160 for training and 15 for validation phases. After evaluating a large number of trials with various MNN architectures, the optimal model was a network with two hidden layers, with 8 neurons in the first and, also, in the second layer, the slabs having the possibility of three activation functions and a jump connection. The mean percentage errors obtained with the best MNN model was of 0.0817%. This result implies that the designed MNN model was properly capable of learning the relationship between the input and output parameters and also confirms that the neural network was able to reproduce simultaneously more that one system, unlike traditional models where one mathematical model was required for each system.
