【作者】 杨敏；

【导师】 江铭炎；

【作者基本信息】 山东大学 ， 通信与信息系统， 2013， 硕士

【摘要】 智能优化算法已随着时代的需求逐渐成为一个成熟的演化计算方法。Hopfield神经网络就作为典型的并行智能优化算法,以其独特的算法优越性而得到了几十年之久的长足发展,现已经成为一种理论与应用研究都较成熟完备的智能优化算法。本论文的主要贡献和创新点如下：介绍了Hopfield神经网络的原理,并针对其在应用中所呈现出的缺陷与不足进行了改进,并将改进后的HNN优化算法应用到通信领域中一些优化问题的解决上。本课题主要提出了三种Hopfield(?)经网络的改进算法,分别为动态步长HNN、模拟退火HNN与动态步长混沌HNN算法。本课题还将改进后的Hopfield神经网络用于解决经典组合优化问题——旅行商问题、认知无线电中最佳频谱资源利用问题以及最大化无线传感器网络的生命周期问题。研究了在认知多载波网络中子载波功率分配算法,并结合Overlay与Underlay两种模型各自的优缺点,提出了一种混合频谱接入算法。本文针对HNN易陷入局部极值的问题,借鉴模拟退火算法的原理,提出了模拟退火-Hopfield神经网络,并将该优化算法应用于实现认知无线电系统的遍历容量最大化问题中。实验结果表明模拟退火-Hopfield神经网络可有效跳出局部极值,提高了取得全局最优解的概率,同时实验表明,混合频谱接入算法可比传统算法取得更好的遍历容量。提出了基于tent映射和动态步长的Hopfield神经网络。用TSP问题验证了此算法的性能优于传统HNN。本文还实现了对传统LEACH分簇协议的改进算法,实现了最优簇头选择与网络生命周期最大化。结果证明此优化算法可以使得簇头的分布更加均匀,并有效延长网络的生命周期,延缓了节点的死亡时间。更多还原

【Abstract】 In recent years, Intelligent Optimization Algorithms has become a mature intelligent optimization algorithm, more and more researchers to join in this field. Hopfield neural networks as a typical intelligent optimization algorithms has been rapid developed because its unique advantage. Now it has been an all aspects relatively complete intelligent optimization technology.This paper introduces the theory of Hopfield neural networks briefly, aim at the shortage in the using of the Hopfield neural networks we proposed many improved algorithms, and then give some applications in the field of communication.The main contributions of this dissertation are as follows:This paper mainly proposed three improved Hopfield neural networks algorithms, which are dynamic step Hopfield neural networks、simulated annealing Hopfield neural networks and dynamic step chaotic Hopfield neural networks. They are effectively in high speed calculating for optimization problems especially for Non-Polynomial problems, and these improved algorithms not only increase the rate of convergence but also improve the veracity of the optimal solutions. This paper also use the improved HNN to solve traveling salesman problem、optimization spectrum using problem in cognitive radio and prolonging the lifetime of the wireless sensor networks.We solve the optimization power allocation problem based on cognitive radio network system. We propose a Hybrid Spectrum Access (HSA) method which considers the total transmit power constraint, the peak power constraint and the primary users’tolerance. In order to solve this combinational optimization problem and achieve the global optimal solution, we derived a Simulated Annealing-Hopfield neural networks (SA-HNN). The simulation results of the optimized ergodic capacity shows that the proposed optimization problem can be solved more efficiently and better by SA-HNN than HNN or Simulated Annealing (SA), and the proposed HSA method by SA-HNN can achieve a better ergodic capacity than the traditional methods.In wireless sensor networks (WSNs), making use of the energy efficiently is becoming increasingly important, we improves the well known cluster-based LEACH(Low-Energy Adaptive Clustering Hierarchy) protocol by defining a new cost function, which aims to minimize the intra-cluster distance and the energy consumption of the network. Moreover, this paper aims at the local optimization and the convergent rate problem of the Hopfield neural networks(HNN), and proposes an improved HNN which is based on dynamic step and tent map chaotic search algorithm and we test the performance of this network by using the TSP. Thus, we solve the optimization protocol by the improved HNN. Our protocol is compared with the traditional LEACH, simulation results demonstrate that the proposed protocol can achieve longer network lifetime.In the end we summarize the characteristics of the application of the algorithm and give future research trends of the Hopfield neural networks.更多还原