Evolving Stable Solutions to Centralized and Distributed Dynamic Constraint Networks Using Hybrid Genetic Search

许多现实世界的问题，如机器人运动规划，调度，无线网络中的频率分配，网络资源分配可以表示为约束满足问题（CSP）的约束网络的基础上。 这些类型的问题通常是分布式的，动态的，并包含大量的变量和约束。 开发能够解决大规模，分布式，动态问题的自适应算法是必要的。 在这一建议中，PI描述了一个研究计划，调查使用混合遗传算法（HGA）不断发展的解决方案，经常性的，动态CSP（RDCSP）。 RDCSP可以被看作是临时添加或删除约束、变量和变量值的CSP。RDCSP可以基于集中式或分布式约束网络。 这项研究将导致对HGA和爬山算法最具挑战性的RDCSPs类型有更深入的了解，并将导致HGA的开发，其搜索效率高于爬山算法

Many real-world problems such as robot motion planning, scheduling, frequency assignment in wireless networks, and network resource allocation can be represented as constraint satisfaction problems (CSPs) that are based on constraint networks. These types of problems are usually distributed, dynamic, and contain large numbers variables and constraints. The development of adaptive algorithms that are capable of solving large-scale, distributed, dynamic problems is a necessity. In this proposal, the PI describes a research program that investigates the use of hybrid genetic algorithms (HGAs) for evolving solutions to recurrent, dynamic CSPs (RDCSP). A RDCSP can be viewed as a CSP where the constraints, variables, and values of variables are temporarily added or removed. RDCSPs can be based on either centralized or distributed constraint networks. This research will lead to a greater understanding of the types of RDCSPs that are most challenging for HGAs and hill-climbing algorithms, and will result in the development of an HGA that searches more efficiently than the hill-climbing