基于改进的GWO-LSSVM磨矿粒度软测量模型  

Soft-Measuring Model for Grinding Particle Size Based on the Improved GWO-LSSVM Model

作　　者：张燕 陈慧丹[1] 周颖 杨鹏 ZHANG Yan;CHEN Hui-dan;ZHOU Ying;YANG Peng(School of Artificial Intelligence and Data Science,Hebei University of Technology,Tianjin 300130,China;Engineering Research Center of Intelligent Rehabilitation and Detecting Technology,Ministry of Education,Tianjin 300130,China)

机构地区：[1]河北工业大学人工智能与数据科学学院,天津300130 [2]智能康复装置与检测技术教育部工程研究中心,天津300130

出　　处：《计算机仿真》2020年第6期298-304,共7页Computer Simulation

基　　金：国家自然科学基金(61773151,61703134);河北省自然科学基金(F2018202279)。

摘　　要：针对磨矿粒度检测周期长且化验过程滞后,难以满足实时在线检测的问题,结合典型二段磨矿回路的特性,采用改进灰狼优化算法(GWO)优化最小二乘支持向量机(LSSVM)磨矿粒度软测量模型,利用改进灰狼优化算法对LSSVM的惩罚系数和核函数参数迭代寻优。针对传统灰狼优化算法求解精度不高、后期收敛速度慢、局部搜索能力弱的缺点,采用自适应位置更新与引入高斯变异进行改进。利用MATLAB仿真得出:改进GWO-LSSVM软测量模型预测精度更高。实验结果表明:改进GWO-LSSVM软测量模型能够效果较好的实现对磨矿粒度的在线检测,以便后期实时控制。Aiming at the problems that the grinding particle size is difficult to meet the online detection, detection cycle is long and the detection process lag, by combining the characteristics of the typical two stage grinding circuits, the soft-measuring model of grinding particle size was proposed based on the combination of improved Grey Wolf Optimization(GWO) and Least Squares Support Vector Machine(LSSVM). The penalty coefficient and kernel function parameters of LSSVM were optimized through the improved Grey Wolf Optimization. Aiming at the shortcomings of the traditional GWO, such as low precision, slow convergence and weak local search ability, an improved GWO based on adaptive search and Gaussian mutation was proposed. MATLAB simulation results show that the improved GWO-LSSVM soft-measuring model has higher prediction accuracy. The experimental results show that the improved GWO-LSSVM soft-measuring model can effectively realize online detection of grinding particle size, so as to facilitate real time control in the later stage.

关 键 词：灰狼优化算法 自适应策略 高斯变异算子 最小二乘支持向量机 磨矿粒度 软测量 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]