进化数据驱动的群体智能算法及其分布式计算模型研究

项目名称： 进化数据驱动的群体智能算法及其分布式计算模型研究

项目编号： No.61472269

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 谭瑛

作者单位： 太原科技大学

项目金额： 80万元

中文摘要： 适应值计算耗时问题是进化类算法应用的瓶颈，也是目前进化算法研究领域的热点问题。高效的适应值函数近似代理模型辅助的进化算法和高性能分布式计算环境是解决此问题的两种有效途径。本项目以微粒群算法为基本算法，同时考虑代理模型辅助的群体智能算法及其分布式计算模型，研究分布式计算环境下的进化数据驱动的群体智能算法。首先，建立历史进化数据的Key-value数据库及其模糊信息粒化模型，构建历史进化数据驱动的群体智能算法的统一集成框架及其分布式计算模型；然后，研究数据驱动的群体智能算法分布式计算模型在分布式计算集群GPGPU环境下高效实现的关键技术，并以典型测试函数为例进行实验研究；最后，以大型起重机的结构优化设计问题为对象开展应用研究。本项目研究可为数据驱动的进化算法及其分布式实现研究提供理论支撑、解决方案和应用案例。

中文关键词： 群体智能算法；代理模型；分布式计算模型；适应值计算耗时问题

英文摘要： Due to the limited cognitive level, mankind has yet to fully understand the internal mechanism of swarm intelligent behavior emergence, which provides a broad space for researching swarm intelligence. In order to explore the emergence mechanism of swarm intelligent behavior based on physics principles, the subject combination of biological and physical systems with self-organizing, self-learning and self-adaptiveness characteristics, bio-aggregative model and foraging model are established inspired by physics laws revealing a regular physical phenomena presented by nature. Through constructing the mapping ralationship between bio-foraging model and swarm optimization algorithm, the subject systemically researches the construction method of environmental indicators and particle state and the mapping ralationship between the two and particle states of matter, and researches how to set the physical properties set of the particle, how to construct the fitness-related properties function, the force law between particles and the property changing function under different particle states of matter. Hence, a unified framework model of swarm intelligence algorithm based on physics principle has been created, which proposes a set of design methods to swarm intelligence algorithm based on physics principle for solving optimization problems, lays the theoretical foundation for researching swarm intelligence from the physics point of view and provides a new ideas for enriching computational intelligence methods.

英文关键词： Swarm Intelligence Algorithm;Surrogate Model;Distributed Computing Model;Computationally Expensive Problem for Fitness Function