项目名称: 变结构学习自动机的自适应与协同优化算法
项目编号: No.61272271
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 张军旗
作者单位: 同济大学
项目金额: 81万元
中文摘要: 学习自动机是一种基于强化学习机制的概率自动机,在与随机环境相互作用的过程中学习未知环境的概率特征。学习自动机自提出以来,其基础理论的研究取得较大发展,形成以变结构学习自动机为主流,并在模式识别、调度与分配、拥塞避免、博弈、马尔科夫链控制、函数优化等众多应用领域取得了显著的实际效果。本项目围绕变结构学习自动机的个体自适应与群体协同两个关键科学问题,首先针对目前变结构学习自动机对不同环境使用固定的结构变化方法并需手动调试参数,使得学习效率与应用范围受限的问题,定义环境的学习难度并提出变结构学习自动机的个体自适应学习范式与追击模式;然后,针对群体变结构自动机通常采用轮询的学习方式,缺乏有效的群体协同机制而导致群体学习能力受限的问题,提出基于群体智能的变结构学习自动机的群体协同算法;最后,针对基于随机反馈的高维多模态函数优化问题建模,发挥具备自适应学习与群体智能的变结构学习自动机的优化能力。
中文关键词: 学习自动机;变结构;自适应;协同;
英文摘要: Learning automaton is a reinforcement learning-based probabilistic automaton used to learn the probabilistic feature of the unknown random environment. Learning automata have been greatly developed in their theory and algorithms. They already gained success in many application fields, such as pattern recognition, scheduling and allocation, congestion avoidance, games, Markov chain control and function optimization. Variable structure learning automata represent the mainstream of the present research. This project concentrates on the self-adaptation and cooperation of the variable structure learning automata. First, we will define the learning difficulty of the environment and propose a self-adaptive learning paradigm and pursuit scheme to resolve the problem of fixed learning algorithm and tuned parameters for different environments. Then, we will propose the swarm intelligence based group cooperation algorithm instead of the commonly used polling approach. Finally, we will model the high-dimensional and multi-modal functions with random feedback using our proposed methods and achieve the optimization goal with adaptive learning and swarm intelligence-based variable structure learning automata.
英文关键词: Learning automata;Variable structure;Self-adaptative;Cooperative;