【作者】 沈洪远；

【导师】 彭小奇；

【作者基本信息】 中南大学 ， 热能工程， 2009， 博士

【摘要】 有色冶金生产涉及经济、国防、航天等多个部门,有色冶金热工过程操作和控制的改善对节能降耗、提高原材料利用率、改善生产环境和降低操作者劳动强度等都有着重要意义。但有色冶金过程因多变量、非线性、大时滞、各变量间强耦合、部分过程参数检测困难、生产过程有时有间歇性等因素而难于操作和控制,目前多数有色冶金热工过程的操作和控制主要靠操作者的经验,而优化的操作规则需要依靠操作者的经验给出。由于受多种因素影响,操作者提供的规则有很大的随意性;另一方面,实际生产过程中记录的大量生产数据一般作为生产运行日志而闲置,这些运行数据中隐含有系统运行规律和操作控制规则。本文研究用数据挖掘的方法,从有色冶金热工过程的生产运行数据中挖掘出系统优化操作和控制的规则,并将其用于铜锍吹炼,仿真结果证明了本文方法的可行性和实用性。本文的主要工作有:1.研究有色冶金热工过程数据挖掘的特点。有色冶金热工过程具有非线性、高维等特点,其数据多为连续量,有较强的噪声,与商业数据挖掘存在明显差异;2.在有色冶金热工过程数据挖掘中引入组件观点,以方便数据挖掘算法的比较和开发。组件观点将所有数据挖掘算法划分为任务、模型、评分函数、搜索方法和数据管理五个组件,使得由各个行业发展出的数据挖掘算法有一个统一的比较和研究框架;3.构建有色冶金热工过程数据挖掘框架,该框架由数据预处理、数据挖掘算法和对挖掘结果的评价构成,并对铜锍吹炼热工过程进行了数据挖掘。铜锍吹炼过程参数呈现多变量、非线性、大噪声的特点,本文应用几种典型的挖掘算法对其成功地进行了数据挖掘,并由此证明数据挖掘理论和技术能有效地应用于有色冶金过程优化决策,实现节能降耗的目标;4.发展了基于改进微粒群(Particle Swarm Optimization,PSO)算法的多峰优化算法。目前多峰优化问题还没有理想的算法,本文提出的算法在低维情况下简单有效;5.提出混沌微粒群优化算法。混沌运动具有遍历性和内在随机性,用混沌序列来产生PSO算法中的初始微粒,使微粒分布更加合理,从而有利于找到优化点;6.提出基于改进的微粒群算法的山峰聚类算法。与现有聚类算法相比,本文提出的算法需主观指定的参数少,聚类效果好;7.提出基于局部微粒群算法的快速山峰聚类算法。与基于微粒群算法的山峰聚类相比,在精度损失很小的情况下,该聚类算法可省去80%以上的计算工作量;8.提出基于PSO山峰聚类的离散化算法。该算法与已有离散化方法相比,需人为确定的参数少,属性值调整方便。更多还原

【Abstract】 The non-ferrous metallurgical production concerns economy, national defense, aerospace engineering and some other fields. The improvement of operation and control of the non-ferrous metallurgical process has a great meaning in saving energy, increasing the raw material utilization ratio, improving the production environment and reducing the operator’s labor intensity. But the non-ferrous metallurgical process is difficult to operate and control because it is often multivariable, nonlinear, largely delayed, strongly coupling, very difficult to measure some process parameters and intermittent in some production process. At present, most of the operation and control in the non-ferrous metallurgical process depend on the operator’s experience. And also the optimized operation rules are given by the operator’s experience. Due to multi-factors, the rules have great casualness. On the other hand, most operation parameters are recorded during the practical production, and these data are regarded as running log and left unused, though the rules of system running and controlling are hidden in those. With the data mining methods employed, the rules of optimized operating and control are extracted from running data of the non-ferrous metallurgical process. Furthermore, the methods are proved to be feasible and practical by the simulation of copper-matt converting.The main work is as follows:1. The characteristics of data mining are studied in non-ferrous metallurgical process. The non-ferrous metallurgical process is generally nonlinear, high dimensional, and the data is mostly continuous and noisy, so it has a great difference from the commerce data mining.2. In order to develop and compare the algorithms, the viewpoint of components is introduced in the data mining of non-ferrous metallurgical process. The view of components divides the algorithms into five modules: task, model, score function, optimization method and data management, thus the data mining algorithms from every area have a uniform research framework.3. The framework of data mining in the non-ferrous metallurgical process is constructed and some examples of data mining in copper-matte converting are given. The parameters of the copper-matte converting process are multivariable, nonlinearity and noisy. In the paper, several data mining algorithms are successfully applied to copper-matte converting. It is proved that the data mining theory and technology can be applied to the optimization decision of non-ferrous metallurgical process, and to achieve the target of saving energy and reducing consumption.4. The multimodal optimization algorithm based on improved particle swarm optimization is developed. At present there is no effective algorithm of multimodal optimization. The algorithm proposed in this paper is simple and effective in low-dimensional.5. The chaos particle swarm optimization algorithm is proposed. Chaos motion has ergodic property and inherent randomness. In PSO algorithm, the initial particles produced by the chaos sequence can be distributed reasonablely, which makes it favorably find the optimization points.6. The mountain clustering algorithm based on improved particle swarm optimization is proposed. Compared with the existing clustering algorithm, the algorithm presented in this paper has fewer determining parameters and the clustering effect is better.7. The fast mountain clustering algorithm based on local particle swarm optimization is proposed. Compared with the mountain clustering algorithm based on particle swarm optimization, the algorithm can save more than 80% amount of calculation with little precision loss.8. The discretization algorithm based on PSO mountain clustering is proposed. Compared with the present other discretization method, determining parameters are fewer and adjusting attribute value is more convenient.更多还原