DefinePK

Abstract

A Dynamic Hybrid Fuzzy Jordan Artificial Neural Network finds the intrusion very efficiently in timely behaviour varied networks. The development of DHFJAN is to influence the behavior of dynamic systems to achieve the pre-determinate objectives. In DHFJAN, the number of hidden layer in neural and Jordan networks and number of nodes in each hidden layers are determined at runtime based on error obtained in the training stage. Generally neural network and Jordan network approaches are inherent nonlinear optimization problem, the quality of Hybrid network local solution is determined the weight initialization. Stability and weight convergence are important issues in the performance analysis of dynamic networks. The weight initialization and optimization of hybrid networks is not discussed in DHFJAN. This paper finds the optimal weight values of each layer by utilizing the optimization techniques to improve the performance and final representation of hybrid network. Many researches were focused on the weight optimization of neural network using various optimization algorithm the PSO is best among them. Since two networks available in hybrid network, applying PSO leads time consumption because the search space size of PSO. This paper proposes a modified PSO algorithm named as CCPSO with constrained search space and controlled convergence degree. The constrained search space is achieved by generating initial weight values based on power-law distribution and Zipf's law. Convergence degree of population in the PSO is controlled by analyzing mean and variance values of fitness in each iteration. The optimal weights are updated in hybrid dynamic network while neuron state changing. Thus the proposed approach improves the performance of dynamic hybrid fuzzy Jordan neural network and also reduces the error rate significantly. Experimental result shows that the proposed WDHFJANN is better than the DIHFJANN.