人工神经网络结构自组织设计方法及应用研究

项目名称： 人工神经网络结构自组织设计方法及应用研究

项目编号： No.60873043

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 生物科学

项目作者： 乔俊飞

作者单位： 北京工业大学

项目金额： 33万元

中文摘要： 人工神经网络是一种旨在模拟人脑结构和功能的信息处理系统，具有广阔的应用前景。神经网络的结构设计是能否复现脑功能的关键，也是神经网络成功应用的核心技术。课题从信息科学的角度出发，借助生物学中的细胞修复原理、条件反射原理和认知科学中的"行动－评价－改进"的原理，拟采用启发式搜索聚类、外科手术法等方法，对人工神经网络拓扑结构的自组织设计算法展开研究。课题主攻的关键问题是建立神经网络性能与神经网络结构元素之间关联模型，获得神经网络结构生长和削减的演化机制，给出保证神经网络结构调整稳定性的动态优化算法等。课题研究目标是使获得的自组织神经网络具有根据承担的任务不同自动调整网络拓扑结构（即增加或删减神经元）的功能。最后，将获得的自组织神经网络应用于污水处理过程控制中，建立其控制模型，争取在实际中推广应用。

中文关键词： 人工神经网络；结构优化；自组织设计；污水处理过程建模

英文摘要： Artificial neural network (ANN) is an information processing system which can simulate the cerebral nerve system (CNS). ANN has been successfully applied in many fields. In fact, the structure design of the ANN is the key technology to reproduce the brain function and the core technologies for the successful application. From the perspective of information science, this project, with biology in cell repairing, conditioned reflex theory and cognitive science "action - evaluation - improvement" principle, is proposed to self-organize the neural network topology by the heuristic clustering method, optimal brain surgeon and so on. The key issues of this project are: modeling the relations between the performances and the structure of the neural network, studying neural network evolutionary mechanisms for growing and pruning, and owning convergent parameters learning algorithms for the dynamic structure neural network. The research aim is to obtain the self-organizing neural networks which can adjust the network topology (ie, adding or deleting neurons) based on the different tasks. Finally, the self-organizing neural networks are used to control the wastewater treatment process, to establish its control model, and other practical applications.

英文关键词： Artificial neural network; structure optimization; self-organizing design; wastewater treatment process modeling