【作者】 王雅娣；

【导师】 曹长修；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 随着最近几年经济的突飞猛进，国内外钢铁市场对钢的质量尤其是纯净钢的质量要求越来越高、越来越苛刻。但是硫的存在却给这带来了很大的难度，因此以质量取胜的钢铁企业纷纷把如何降低钢材中的硫含量、提高钢材质量和开发新品种视为钢铁生产的一个核心战略任务，以增强企业的市场竞争力。 然而目前广泛应用的铁水预脱硫主要采用人工控制，这导致对生产过程的判断和操作调剂的主观性很大，难以保证脱硫操作的稳定性，导致钢产品质量波动大，脱硫成本增加。本文以大型钢铁生产基地攀钢集团为背景，采用神经网络来建立脱硫静态模型，自动寻找脱硫过程的规律和知识，从而对脱硫过程进行决策支持，降低脱硫成本，为全自动脱硫创造了良好的条件。 本文采用RBF神经网络作为建模工具。针对建模过程中出现的RBF中心和宽度难以确定的难点，在分析蚂蚁算法机理的基础上，提出了使用智能蚂蚁算法对RBF神经网络模型的中心和宽度进行自适应选择，从而达到模型训练精度和范化能力的一个最优的平衡，从而提高模型的预报精度。 文中首先详细介绍了基本蚂蚁算法的思想和特点，然后在分析其发展现状和局限性的基础上，采众家之长，决定采用智能蚂蚁算法来优化RBF神经网络的中心和宽度。文中针对基本蚂蚁算法容易出现停滞、参数难以确定的局限性，对其进行了一定的改进——智能蚂蚁算法：（1）引入蚁群优化算法中对转移概率公式、信息素更新规则的修改；（2）引入Max-Min蚂蚁算法中对轨迹强度τ_ij设置上下限τ_max和τ_min；（3）在蚂蚁算法中加入局部优化，从而进一步缩短解路线的长度，以加快蚂蚁算法的收敛速度；（4）对参数进行了一定的选择。最后通过程序仿真证明了智能蚂蚁算法与基本蚂蚁算法相比具有明显的优越性。 本文在分析脱硫工艺原理的基础上，通过有效的数据预处理，最后进行仿真分析，基于智能蚂蚁算法优化的脱硫静态模型与传统的RBF神经网络脱硫静态模型相比较，其模型的预报精度好于传统脱硫静态模型，具有一定的实用性和推广性。更多还原

【Abstract】 With the rapid development of economy, especially recent years, the domestic and overseas markets of steel-iron require the quality of steel, in particular the pure steel, higher and higher. However, the existence of sulfur brings about the large hidden trouble. So, it is a key strategy for many steel-iron-making enterprises which get victory relying on the quality to remove the sulfur efficiently, improve the quality of steel and design the new products in order to win more share in steel-iron market.However, presently the popular iron water pre-desulfuration method is mostly based on manual control which is too subjective to make a right decision for process and manipulation sometimes, which results in the instability of the process, harms the quality of steel and increases the production cost. On the background of panzhihua steel-iron corporation which is one of the big national steel-iron corporation, a static model of desulfuration is designed in this dissertation. The modeling process is based on Neural Network, which can automatically discover the rules and knowledge in desulfuration process, and then provide the decision support, improve the desulfuration effect and decrease the cost. The successful use of this static model will be a solid prerequisite for full automatic dusulfuration in the future.The Radial Based Function (RBF) neural network is adopted as the modeling algorithm. To overcome the difficulty in determining the RBF center numbers and spread, a kind of Intelligent Ant Algorithm is introduced, which follows the analysis of the basic rules of ant algorithm. The new hybrid algorithm determines the center numbers and spread adaptively to reach the optimal balance between the training accuracy and the generalization, so it increases the prediction accuracy of the model.Firstly the dissertation explains the ideas and characters of Ant Algorithm. Then on the basis of analyzing its development and limitation, Intelligent ant algorithm adding many advantages of some improved ant algorithms is adopted to optimize the center numbers and spread of RBF Neural Network. Aiming at the limitation of standard ant algorithm, this dissertation improve them by some modification: (1) The formula of transfer probability and the rule of pheromone updating is modified, i.e. Ant Colony Optimization Algorithm; (2)The intensity of trail ij is restricted between min and max , i,e- Max-Min Ant Algorithm; (3)Adding local optimization into ant algorithm to improve the most optimal path in every generation in order to shorten the optimal path and to quicken the speed of convergence; (4) analyzing the choice of some mainparameters. Finally the experiment results prove that Intelligent Ant Algorithm is obviously better than standard ant algorithm.After analyzing the desulfuration techniques, through the effective data preprocessing, this dissertation adopts Intelligent Ant Algorithm to optimize the desulfuration static model. Comparing the performance between the improved model and traditional RBF model by simulation, it proves that the former is better than the latter and has its own practicability and validity.更多还原