【作者】 周轩；

【导师】 钱锋；

【作者基本信息】 华东理工大学 ， 模式识别与智能系统， 2011， 硕士

【摘要】 生物免疫系统是一个高度进化的生物系统,从计算的角度来看,生物免疫系统是一个高度并行、分布、自适应和自组织的系统,具有很强的学习、识别、记忆和特征提取能力。人们基于生物免疫系统开发了人工免疫系统,借鉴免疫系统的功能和原理用于解决复杂问题。本文在深入研究人工免疫系统的基础上,分析了免疫系统在增量学习方面的优势,将其用于弥补PSO算法该方面的缺点,然后又讨论了免疫聚类算法设计模型,最后分析了免疫优化系统目前存在的缺点,并提出了有效的解决办法。本文首先对数据分类进行了探讨。当前各个领域中的目标数据越来越庞大,对算法增量学习能力的要求也越来越高。免疫记忆机制作为AIRS增量学习能力的核心,可为其他算法所用。本文针对PSO算法在增量学习方面的弊端,引入AIRS的免疫记忆机制,开发了一种免疫混合PSO增量分类算法。通过在若干UCI标准数据集上进行的仿真,证明该混合算法具有增量学习能力,且在分类准确率方面也比一些经典分类方法具有优势。其次分析了聚类问题中的免疫方法,针对目前免疫聚类方法繁多,却没有一个系统的算法设计框架的问题,本文通过对一个成功的免疫聚类算法—aiNet的深入研究和归纳总结,提出了一个系统的免疫聚类算法设计框架。该设计框架将聚类算法的设计分割成了五个主要部分,并对每个部分采用的设计思想进行了细致的阐述,为新算法的设计给出了框架上的指导。最后深入探讨了免疫算法在优化问题中的应用,针对opt-aiNet算法种群无规律变异导致其在单目标优化中收敛速度降低的问题,本文对opt-aiNet算法抗体的变异过程进行了改进,引入了PSO算法中粒子飞行的策略,使抗体具有了有方向变异的能力,提出了一种具有双变异机制的opt-aiNet算法。本文还将新算法应用于无线网络规划问题,与无线网络规划中普遍应用的GA相比能更快更好地发现全局最优解,同时和传统opt-aiNet相比在搜索效率上有了较大提高。更多还原

【Abstract】 Artificial immune system (AIS) was developed based on biological immune system which is a highly parallel, distributed, adaptive and self-organizing system, and has a strong learning, recognition, memory,,and feature extraction capabilities. In this dissertation an immune memory strategy was used to make up the weakness of PSO. Then a framework for algorithm designing was discussed. Furthermore a shortcoming of immune optimization algorithm was studied, and an effective solution was proposed. The main contributions of this dissertation are described as follows:Firstly, classification problem was discussed. Due to the scale of target data for classification task was growing larger, a incremental learning technology was required. In order to make up the drawbacks of PSO in incremental learning abilities, the AIRS memory strategy was injected, and a hybird algorithm--AIR-PSO was presentd. Later AIR-PSO was tested on a number of UCI standard data sets. It showed that the hybrid algorithm had the ability of incremental learning, and also had the advantage in classification accuracy.Secondly, a general framework for immune clustering algorithm was discussed. Based on a depth study on aiNet process, a general framework for immune clustering algorithms was proposed. The framework was build up by five parts, and each part adopted a lot of crucial methods. The framework gave a systematic guidance for the design of new algorithms.Finally, the immune optimization algorithm was studied. Opt-aiNet algorithm encountered a lower convergence rate in single objective problems sometimes. Studies have shown that the random mutation of the population may lead to this problem. Then an oriented mutaion process inspired by PSO evolution strategies was proposed, and a bi-mutaion opt-aiNet was raised up. After that the bi-mutaion opt-aiNet was applied to the wireless network planning, and achieved a better result than GA and the traditional opt-aiNet.更多还原