Types Passive Analysis Of Neural Networks With Delay

Because Neural Network has those distinguished capabilities like distributive information storage, self-organization, self-learning and parallel processing, it's a very powerful tool in optimization, fault diagnosis, associative memory, pattern identification, being widely used in industrial control, clinical medicine, risk alert for commercial bank loans, etc. However, it is necessary to analyze NN with complex structure for a good grasp of its complex dynamic behavior including stability and passivity so as to fully develop its great potential.This paper first does an analysis on the passivity of Cellular Neural Network (CNN) with varying time delay, and further obtained sufficient condition for global robust passivity of CNN when considering uncertain parameters in the system. The introduction of delay segmentation approach in the paper makes derived condition less conservative than the conditions in other articles.Then, passivity of CNN with distributive delay and infinite delay is discussed, and by proper inequality scaling and addition of weighting matrix into Lyapunov function, constraints on delay function is relaxed. At last, a class of CNN having multi-time-varying delay and neutral delay are studied, and general condition for passivity of network with infinite number of delays is derived that also applies to single varying delay system.The analytical approach is mainly based on Lyapunov-Krasovskii stability criterion, and better passivity criterion can be obtained with inequality transformation method and through the techniques of using Lyapunov function with delay segmentation, introducing more slack variables, considering usually neglected or exaggerated item.Besides, the delay functions in the network model and the corresponding neural connectivity functions are not limited to be identical, making the results more general. Finally, Schur theorem is used to derive robust passivity condition for the system in LMI form. All the conditions gained in the paper are simulated in MATLAB to test their feasibility.