【作者】 刘元东；

【导师】 王海东；

【作者基本信息】 中南大学 ， 材料学， 2004， 硕士

【摘要】 水泥熟料的煅烧是一个非线性、多变量、大滞后、强干扰的复杂工业过程，进行水泥熟料煅烧质量的预报对于实现水泥生产过程的优化控制具有重要的作用。由于传统的预报方法难以满足这种复杂系统的预报建模要求，因此，应用人工智能技术，建立水泥熟料煅烧质量的人工神经网络预报系统也就具有重要的意义。 通过对水泥熟料煅烧系统进行分析，提出了由数据预处理系统和神经网络系统组成的水泥熟料煅烧质量的神经网络预报系统方案；研究了神经网络建模所需的数据预处理方法，包括数据清洗、数据变换和神经网络变量选取方法；研究了BP网络模型和BP算法，并对BP算法进行了加入动量项和自适应调节步长的改进。 构筑了一个具有通用性的开发复杂软件的系统框架，应用此软件框架开发了神经网络建模工具软件，进而建立了水泥熟料煅烧质量的神经网络预报模型，并得了较好的预报结果：立升重的预报相对误差为1.8％；f-CaO含量的预报相对误差为2.8％。更多还原

【Abstract】 The cement clinker calcination is a complicated industry process which is nonlinear, greatly lagged, multi-variable and strongly disturbed. The forecast of cement clinker calcinations quality is very important to the optimization control of the produce process of cement. Because the require of forecast modeling for this complicated system is difficult to be satisfied by the traditional forecast methods, it is significant to apply artificial technology to build the neural network forecast system for the calcination quality of cement clinker.Through analyzing the calcination system of cement clinker, the scheme of neural network forecast, which include neural network system and data pretreatment system, is proposed; The method of data pretreatment, which is needed for neural network modeling and consist of cleaning data, switching data and selecting the variables of neural network, is researched; BP network model and BP algorithm are also researched and BP algorithm is improved by adding the momentum item and self-adapting the length of train step.A general system frame which can be used to develop complex software is built. A tool software for modeling neural network is developed on this software frame. A neural network forecast model of cement clinker calcination quality is built by this tool software. A good forecast result is got through this model: the relative error of g/L is 1.8%, the relative error of the content of f-CaO is 2.8%.更多还原