自动波竞争神经网络及其在单源最短路问题中的应用

将竞争机理引入网络的自动波产生与传播过程中，提出自动波竞争神经网络(ACNN)模型，并成功地应用于求解单源最短路问题，给出了基于ACNN的最短路求解算法. 与其他神经网络最短路算法相比，基于ACNN的最短路算法具有网络所需的神经元数目少、神经元及网络的结构简单、易于软硬件的实现、以及全并行方式计算等特点，可用于求解非对称赋权图的最短路径树问题，且其计算时间(迭代次数)仅正比于最短路径上的跃点数，与赋权图的复杂度、路径总长、边长的精度、通路总数等因素无关. 计算机仿真结果表明该算法的有效性和快速求解能力. 关键词： 单源最短路问题 自动波竞争神经网络 脉冲耦合神经网络

Autowave-competition neural network and its application to the single-source shortest-paths problem

In this paper, the competitive mechanism is introduced to the production and propagation processes of the autowave of neural network. The autowave-competition neural network (ACNN) is proposed to successfully resolve the problem of single-source shortest paths (SSSP). The algorithm for shortest paths based on ACNN is presented. Compared with other neural network based approaches, the new algorithm has the following advantages: less number of neurons needed, simple structure of neurons and networks, readily available software and hardware. When ACNN is employed to resolve the shortest path problem, the computational complexity is only related to the hop number of the shortest path, but independent of the complexity of path graph, the number of the existing paths in the graph and the precision of the length of edges. Simulations show that the proposed algorithm is plausible and effictive.