Simulated Annealing Strategy in Chaotic Neural Network with Legendre Function
Yaoqun Xu, Zhenhua Yang, Xinxin Zhen · 2021
As to the monotonous activation funcation of chaotic neural network, a new transient chaotic neuron model is constructed by combining multiple Legendre polynomials and sigmoid functions as a new non-monotonic excitation function, and the dynamic characteristics of chaotic neurons are analyzed. A new transient chaotic neuron network model is constructed by using this model. This model improves the ability of the network to fall into local minima by using the idea of piecewise simulated annealing. The effectiveness of this model has been verified by the application of multi-user detectors in nonlinear function optimization, traveling salesman problem (TSP) and direct sequence-code division multiple access (DS-CDMA) on multi-user detector.