【作者】 李颖；

【导师】 张邦礼；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 上世纪中叶以来,微电子、计算机、通讯、网络、信息、自动化等科学技术的迅猛发展,掀起了以信息技术为核心的“第三次浪潮”,正牵引着人类进入工业经济时代最鼎盛的时期,并打开了知识经济时代的大门。随着计算机技术的飞速发展和应用的普及,人类社会己经进入了一个信息化的时代,人们利用信息技术生产和搜集数据的能力大幅度提高,使用数据挖掘技术对对象建模并进行预测成为发展方向。文中以国内某大型钢铁公司的高炉生产数据为背景,以铁水中硅含量为主要的预报依据。由于高炉的热状态的输入输出数据集间存在着时间上的关系——这类数据称为时态数据(Temporal Data)。所以在对时态数据进行数据挖掘的过程中,必须考虑数据集之中数据间存在着的时间关系。本文对基于时态数据挖掘方法进行了分析,提出了使用RRBF神经网络建立高炉的热状态预测模型。RRBF神经网络模型是在RBF神经网络的输入上加入了自反馈的神经元,使RRBF网络对过去时态的数据具有了记忆能力,其学习算法为在线的学习算法。通过RRBF神经网络模型预报铁水中硅的含量以达到预报高炉热状态的目的。为提高高炉热状态的预报精度,稳定钢铁质量,稳定生产工艺创造了良好的条件。文中详细介绍了利用人工神经网络建立预测模型的思想及其特点,从生物角度和数学推理的方面阐述了神经网络的工作方式。分析了RBF及RRBF神经网络的网络结构及训练算法。最后本文用Matlab建立做为仿真平台,建立了传统的RBF网络和RRBF网络的仿真模型。通过测试,RRBF网络的预报精度和训练速度都明显好于传统的RBF神经网络模型,可见RRBF在对非线性时间序列上数据挖掘中具有明显的优越性。更多还原

【Abstract】 With the rapid development of micro-electronics, computer, communication, Internet, message, automation technologies since the middle period of 20th century. The third tidal wave centered on message technology has come, which results in the best flourishing period of industry economic era, and opens the door of knowledge economic era. Human society has entered into the information era with the development and popularization of computer technology. The capacity of utilizing information to manufacture and collect data has improved greatly. Data mining technology has been the direction of prediction and establishing model.The Silicon content in the molten iron is regarded as a major prediction object based on manufacturing data of blast furnace. We must take the time relationship into account because there is a relationship between the input and output data sets(Temporal Data), which RRBF is used to establish the heat state prediction model in the paper. The self-connection neural unit has been added to RBF NN which made RBF NN had the capacity of memorizing the past-time data . The on-line learning method is also adopted in this paper. By predicting the content of silicon in molten iron the heat state of blast furnace is gained, which provide the fine condition to improve quality of steel and stabilize manufacturing craftwork.The ideas, features and operation of artifical neural network are introduced in detail. from the biology point and mathematics reasoning aspect, then it analyzes the architecture and training algorithm of RBF and RRBF. Finally, through simulation using Matlab, the dissertation proves that the RRBF network is better than traditional RBF neural network model in prediction precision and speed of training. Thus it can be seen that RRBF has obvious superiority in analyzing non- linear time-series data.更多还原