一种新的基于广义细胞自动机的网络快速包交换优化方法

实时优化求解快速包交换问题(FPS)是提高网络性能的重要手段．基于梯度下降法等数学规划方法，不能并行地实时地优化求解FPS问题，而基于Hopfield型神经网络和细胞神经网络的优化方法中，都只有单一粒度的细胞动力学方程和单一粒度细胞之间的相互作用，不仅收敛到平衡点的过程长，而且神经网络参数的选择和修正十分困难．该文提出一种新的具有多粒度宏细胞的广义细胞自动机模型和方法，广义细胞自动机中的小粒度宏细胞聚合成可以独立演化的大粒度宏细胞，通过多粒度群体的不同程度群体智能的相互作用，能够比目前其他方法更快更有效地分布并行地优化求解FPS问题和其它类似的复杂的网络优化问题．

A New Optimization Approach Based on Generalized Cellular Automata for Fast Packet Switching in Computer Networks

The parallel real-time optimization of fast packet switching (FPS) in computer networks is of great significance for improving network performance. The problem is very difficult to solve in real time by using the conventional mathematic programming based on sequential gradient-descent method. On the other hand, although the Hopfield-type neural networks (HNN) and cellular neural networks (CNN) have been widely applied for optimization problem solving, nevertheless there still exist some formidable difficulties with these methods, e.g., they have to take a relatively long time to converge to a feasible solution, and have to determine by experiment or by experience a variety of neural network parameters. Both HNN and CNN share a common hallmark that their neurons/cells only have a single granularity. By contrast, this paper presents a new generalized cellular automata approach (GCA) to FPS problem solving, which is featured by the architecture and evolutionary dynamics of multi-granularity multi-layer macro-cells. The GCA approach, hence, can take advantages of colony intelligence of multi-granularity macro-cell colonies, and exhibits superiority over other presently used optimization methods for FPS problem solving in terms of the parallelism, real-time performance, optimal degree of solution, easiness to determine parameters, and feasibility of hardware implementation with VLSI systolic array.Besides FPS, the proposed GCA approach can also be used to solve a class of network communication problems.