利用极值蚁群优化的制粉出力建模变量选择算法  

Variable Selection Method Based on Extreme Ant Colony Optimization for Pulverizing Capability Model

作　　者：贾立新[1] 郑伊飞[1] 曹晖[1] 

机构地区：[1]西安交通大学电力设备电气绝缘国家重点实验室,西安710049

出　　处：《西安交通大学学报》2011年第10期7-12,共6页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(61005058)

摘　　要：针对制粉出力难以直接测量,以及制粉系统包含变量多且各变量间耦合性强的问题,提出了一种利用极值蚁群优化的制粉出力建模变量选择法,并采用支持向量机根据所选变量建立了制粉出力的预测模型.该算法基于蚁群优化的正反馈原理,对蚂蚁搜索到的各个变量的相对重要性加以区分,并根据幂律分布选择重要性较小的变量进行变异,使得较差解不断得到改善,从而引导蚂蚁朝着最优解的方向搜索.采用制粉系统现场数据对所提算法、蚁群算法和蚁群遗传算法进行比较,结果表明,所提算法具有更快的收敛速度,且由其所选变量建立的制粉出力模型具有较高的预测精度.Since the pulverizing capability is unable to be measured directly and there exist numerous strongly-coupling variables in a ball mill pulverizing system, a variable selection method based on extreme ant colony optimization （EACO） for pulverizing capability model is proposed. According to the selected variables, the model is established by support vector regression. On the basis of the positive feedback mechanism of ant colony optimization （ACO）, the proposed method is able to identify the relative importance of variables selected by ants, and then mutate unimportant variables following the power law distribution, thus the poor solutions in each iteration process get improved. Consequently, the ants are led to search for the optimal solutions. The proposed algorithm is compared with ACO and ant colony optimization with genetic algorithm on the field data of hall mill pulverizing system. The experiment results show the higher convergence rate and prediction accuracy.

关 键 词：制粉系统 制粉出力 变量选择 极值蚁群优化 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]