一种径向基混沌神经网络的分段退火策略

摘要

采用Sigmoid与一种径向基函数即逆多二次函数加和形式作为激励函数,分析了该新型混沌神经网络神经元模型的动力学特性,研究了分段线性模拟退火策略,合理改善了模拟退火参数取值不宜过大或过小的问题;利用分段收敛方案改进了该新型径向基混沌神经网络,并将其应用于解决TSP（旅行商最短路径问题）,改进后的网络能够保证合法路径比例的同时,可以以较快的速度收敛到最优解,由此说明此方案的可行性及有效性。%The activation function was composed by Sigmoid function and one radial basis function named Contrary Multiquadric function.The dynamics characters of the neuron model was analyzed and the partitioned simulated annealing strategy was studied to solve the problem about the value of simulated annealing parameter neither too large nor too small.The novel chaotic neural network with radial basis function was improved by the partitioned convergence proposal and was used to solve Traveling Salesman Problem（TSP）.The simulation results indicated that it could ensure both the rate of legitimate path and the speed of convergence and prove the feasibility and the effectiveness of this proposal.