【作者】 王康平；

【导师】 周春光；

【作者基本信息】 吉林大学 ， 计算机应用技术， 2008， 博士

【摘要】 群体智能是基于自治、非中央控制系统的聚合行为的人工智能。随着对复杂系统的深入研究,传统的“从上而下”的研究方法遇到了很多的困难。一种新的研究方法---群体智能被广泛应用在复杂系统的优化、模拟、仿真等领域。群体智能方法两个重要的研究领域是优化模型和基于agent的模拟上。在优化模型中比较著名的是蚂蚁算法和粒子群算法。人工社会系统是基于agent的模拟的重要组成部分。本文主要研究内容如下:1)介绍了群体智能方法的起源及基本方法学。2)介绍了基于群体智能的优化算法,其中包括蚂蚁算法,粒子群算法和基于Agent的优化模型。3)介绍了群体智能方法在人工社会系统中的应用。基于agent的建模方法在人工社会系统中得到了广泛的应用。同时,群体智能方法提供了一种研究社会系统的新方法—模拟。4)引入一种特殊的算子以使粒子群算法可以用于求解旅行商问题。5)使用多Agent模型求解多目标优化问题。在模型中,每个agent都有自己的信念和状态,通过信息的交流和学习自适应的搜索解空间。每个agent都有不同的偏好来优化不同的目标。6)研究了基于agent的拍卖市场。提出了一个包括生产过程的经济模型,研究了在这个模型中一些经济现象。7)使用元胞自动机模型建立的微观交通模型研究交通信号控制策略的性能,研究了不确定交通信息下各种控制算法的效果,同时研究了双环路中同步控制策略的特性。本文的研究结果丰富了群体智能方法在优化和模拟方面的内容,加深了对群体智能方法的认识,有一定的理论意义和应用价值,为进一步的研究提供帮助。更多还原

【Abstract】 Swarm intelligence (SI) is artificial intelligence based on the collective behavior of decentralized, self-organized systems. It is inspired by nature swarm behaviors. The examples of nature SI include ant colonies, bird flocking, animal herding, bacterial growth, and fish schooling. The expression was introduced by Gerardo Beni and Jing Wang in 1989, in the context of cellular robotic systems. The traditional“up-down”methods encounter lots of handicaps in the research of complex systems. A new method, swarm intelligence which is a“bottom-top”research method, is exploited widely in optimization and simulation of complex systems now. Swarm intelligence technology is particularly attractive because it is simple, effective and robust. Swarm intelligence focuses on self-organized, decentralized methods. The swarm systems typically are made up of population of simple agents which interact with each other and environment. The macrostructures emerge from the interactions and operations of agents despite the simple micro rules.There are many optimization models based on SI, and particle swarm optimization algorithm is a typical one. It is inspired by the social behavior of bird flocking. In this algorithm, every particle will modify its search direction and speed according to the best swarm solution and its own best solution. Following this rule, the swarm can search the solution space effectively. The micro-structure of PSO algorithm is simple, but the macro searching behavior is emerging from it based on inter-operations of particles.The traditional macro models cannot explain many social phenomena clearly. In such system, there are lot of agents which have interactions and inter-influences, e.g. traffic systems and economic systems. The micro-models based on swarm intelligence approach have made great improvements in recent years. Most of these models focus on sociology and economics. The agent-based computational economics is a classical one. SI presents a different perspective from micro level. Swarm intelligence techniques have lots of advantages to simulate these systems. In simulations of economic systems, swarm intelligence is used to study and validate economic hypotheses that are hard for traditional approaches.The work of this dissertation focuses on optimization and simulation applying SI approaches. There are two important issues in the SI optimization model: How to map agents to solution space? How agents cooperate to improve the searching efficiency.The two main parts of the work on optimization are listed below:1) A new PSO algorithm is presented for solving travel salesman problem. Swap operator is presented as“speed”to change the current solution to a new one. The distance of two solutions can also be denoted as“swap operator”. In this way, particle swarm optimization can be used in discrete domain.2) An agent-based algorithm is presented for solving multi-objective optimization problems. In multi-objective problems, the key is how to mediate different objectives. Agents with different intents are used to search the solution space. Agents can exchange information with their neighbors in the searching process to improve the local searching efficiency. One evolutional operator is also employed. The experimental results show this method is effective.In simulation area, the following subjects are studied:1) We research CDA market. Some well-known strategies are implemented to study their performances.2) Agent-based simulation on pricing in the barter market. The continuous double auction model in which the supply and demand are fixed is used to explore the pricing process of agents. In this dissertation, a novel and interesting agent model is proposed to explore some properties of a typical and meaningful phenomenon in economics, pricing.In our model, not only auction is included, but also production. We analyse the pricing procedure. We find the price is stable and the variance of price is small. There are some macro phenomena emerging from this model, e.g. the business man, the social labor allocation. And then the impact of transaction capability is researched.3) Traffic signal simulation and research. A traffic system is simulated on cellular automata. Because of the complexity of environment, the traffic information always contains some noises. In this model, some traffic signal methods are tested with noisy traffic information. We find the complicated intelligent strategies are worse when there are noises in traffic information. The light control methods depend on the road structure and the traffic density severely. Traffic light synchronization method is studied on double ring road structure with five different densities. With one of the densities, the traffic light synchronization method has some problems. An improved method is presented to handle them, and the results are promising.In the end, the summaries and further works are presented.更多还原