【作者】 冯登超；

【导师】 杨兆选；

【作者基本信息】 天津大学 ， 信号与信息处理， 2008， 博士

【摘要】 复杂性科学是21世纪一门新兴的边缘、交叉学科,探索复杂性正在成为当代科学最具革命性的前沿。论文工作深入地研究了蚁群算法与小波网络及其改进算法,探索了在复杂性科学中的应用,包括群集智能中的蚁群算法理论、蚁群算法的参数优化设置方法、遗传蚁群算法的改进及其在植物病斑检测中的应用、小波网络的初始化参数设置及基于蚁群算法的小波网络结构优化方法、基于改进型小波网络的决策级信息融合模型的构建。论文的创新点体现在以下四个方面:(1)提出了元启发框架下蚁群算法的参数设置原则及基于正交试验设计方案的参数优化设置方法。对蚁群算法的主要参数采用统计分析方法进行相关性分析,利用正交试验设计减少参数设置的试验次数,实现了最佳参数组合方案,克服了参数设置过程中的主观性,提高了参数选择的效率。(2)提出了一种自适应遗传蚁群算法。分析了遗传算法和蚁群算法的融合策略,研究了遗传蚁群算法中交叉率和变异率的自适应选取算法,并采用自适应信息素挥发因子实现信息素的动态更新。最后,根据植物病斑图像特点,研究了遗传蚁群算法中信息素更新函数和启发函数的改进方法,从全局组合优化角度实现了植物病斑检测。(3)提出了基于蚁群优化的小波网络。构建了基于蚁群优化的小波网络学习算法,利用蚁群算法的全局优化能力实现了对小波网络的权值、阈值、尺度因子、平移因子的优化设置。(4)提出了一种基于改进型小波网络的决策级信息融合模型。在深入研究数据缺失机制及其处理方法的基础上,将多个小波神经网络并行连接实现了基于小波网络的特征级信息融合模型,再结合证据理论构建了数据缺失机制下的决策级信息融合模型。仿真实验验证了所提出的改进算法和信息融合模型的正确性。更多还原

【Abstract】 Complexity science is a new developing interdisciplinary in the 21st Century. Exploring complexity is becoming the frontier of modern science. This paper mainly focuses on the improved algorithm of ant colony algorithm and wavelet network,explores the application in complexity science correspondingly, which includes the theory analysis of ant colony algorithm based on swarm intelligence, the parameters optimization setting of ant colony algorithm, the improved ant colony algorithm based on genetic algorithm and its application of plant disease spot detection,the initialization setting for the parameters of ant colony algorithm,the structural optimization of wavelet network based on ant colony algorithm and the model construction of decision level information fusion algorithm based on improved wavelet network.The innovations of the thesis are embodied in four aspects as follows:(1)The principle of parameters setting for ant colony algorithm based on meta-heuristic frame is proposed and the orthogonal experiment method is designed to optimize the parameters setting as well. The correlation analysis of key parameters can be used by statistical method, the experiment times be reduced by orthogonal design, the optimal settings be obtained, the subjective problem be overcome and the efficiency of parameter selection be improved correspondingly.(2)Adaptive genetic ant colony algorithm is proposed and the detection model of plant disease spot is designed by the global combination optimization method based on adaptive genetic ant colony algorithm correspondingly. In the adaptive genetic ant colony algorithm, the submodule of genetic algorithm and ant colony algorithm are improved respectively. As for the submodule of genetic algorithm, the adaptive selection algorithm of crossover probability and adaptive mutation probability is designed and the selection principle of fitness function is discussed, too. For the submodule of ant colony algorithm, the dynamic pheromone updating mechanism based on adaptive pheromone volatilization factor is designed,which can realize the final improvement of genetic ant colony algorithm. According to the principle of global combination optimization, the plant disease plot detection based on improved genetic ant colony algorithm is finally realized. (3)Improved wavelet network based on ant colony algorithm is proposed.According to the global optimization capacity of ant colony algorithm,the learning algorithm of wavelet network based on ant colony algorithm is constructed to realize the optimization setting of weight, threshold, scale factor and translation factor.(4)The model construction of decision level information fusion algorithm based on improved wavelet network is proposed. On the basis of data missing mechanism and its processing method, a series of wavelet network with parallel network structure are combined to construct the feature level information fusion model.Finally, the model construction of decision level information fusion algorithm is designed by wavelet network and evidence theory.Simulation results verified the correctness of the above improved algorithm and information fusion model.更多还原