Abstract:

To prevent the network from being trapped in the local minima effectively, a novel chaotic neural network with scale annealing of wavelet and hysteretic activation function is proposed. Gauss wavelet function is used for the selffeedback of the network, and chaotic simulated annealing is realized by the exponentially decaying wavelet scale, which enable the network to exhibit more abundant chaotic dynamic behavior and enhance the average level of Lyapunov exponents of chaotic search effectively. Both the optimization property and the stability are analyzed by applying the unified framework theory. The simulation results on traveling salesman problem (TSP) and direct sequencecode division multiple access (DSCDMA) suggest that the network can find global optimal solutions of optimization problems, and it has superior optimization performance.