作者：罗小燕, 陈慧明, 卢小江, 熊洋
作者单位：江西理工大学机电工程学院, 江西 赣州 341000
关键词：磨机负荷;网格搜索;交叉验证;参数优化;混合编程
摘要：
针对实际生产中只能依据专家经验判断磨机负荷（ML）状态，难以检测ML及其直接相关参数的问题，该文通过分析反应磨机内部负荷的振动信号，提取频谱特征，利用支持向量机（SVM）建立磨机负荷参数的预测模型。为解决SVM核函数参数g和惩罚因子C主观选取问题，提出采用网格搜索与交叉验证相结合的方法，对SVM参数进行优化。最后基于Matlab与VC混合编程，建立仿真平台，实现球磨机负荷参数的预测仿真。分别利用SVM默认参数和最佳优化参数代入ML回归预测模型，通过参数ML预测对比，得到SVM最佳优化参数下训练的平均平方误差（MSE）和平方相关系数（r2）均优于SVM默认参数下的预测结果。
Forecast of SVM mill load based on grid search and cross validation
LUO Xiaoyan, CHEN Huiming, LU Xiaojiang, XIONG Yang
School of Mechanical and Electrical Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, China
Abstract: In order to address the problem that mill load (ML) state only can be estimated with expert experience and ML parameters directly relevant to ML and ML state are hard to be monitored during actual production. This article proposed the method of combining grid search and cross validation to optimize SVM parameters and solve the problem of subjective selection between SVM kernel function parameter g and penalty factor C, by analyzing and reflecting vibration signal inside mill, extracting spectrum signature of signal and establishing forecasting model for mill load parameters by using support vector machine(SVM). Forecast simulation of ball mill load parameter was achieved by establishing simulation platform based on hybrid programming of Matlab and VC. Both training mean square error(MSE) and square correlation coefficient(r2) obtained under optimized SVM parameters are higher than the forecasting results under default SVM parameters by respectively substituting default SVM parameters and optimized SVM parameters into ML regression forecasting model and ML parameter forecasting and comparison.
Keywords: mill load;grid search;cross validation;parameter optimization;hybrid programming
2017, 43(1): 132-135144 收稿日期: 2016-06-22;收到修改稿日期: 2016-07-20
基金项目: 国家自然科学基金项目（51464017）；江西省高等学校科技落地计划项目（KJLD1304（45））
作者简介: 罗小燕(1967-),女,江西赣州市人,副教授,硕士生导师,研究方向为机电系统智能监测与控制。
参考文献
[1] 石立,张国旺,肖骁. 金属矿山选矿厂磨矿分级自动控制研究现状[J]. 金属材料与冶金工程,2014(1):43-48.
[2] WEI D H, CRAIG I K. Grinding mill circuits a survey of control and economic concerns[J]. International Journal of Mineral Process,2009,90(14):55-56.
[3] HUANG P, JIA M P, ZHONG B L. Investigation on measuring the fill level of an industrial ball mill based on the vibration characteristics of the mill shell[J]. Minerals Engineering,2009,22(14):1200-1208.
[4] TANG J, ZHAO L J, ZHOU J W, et al. Experimental analysis of wet mill load based on vibration signals of laboratory scale ball mill shell[J]. Minerals Engineering,2010,23(9):720-730.
[5] 冯雪,赵立杰,郑瀚洋. 基于KPCA频谱特征提取的球磨机负荷检测方法[J]. 沈阳化工大学学报,2014,28(2):175-180.
[6] 汤健,赵立杰,岳恒,等. 基于多源数据特征融合的球磨机负荷软测量[J]. 浙江大学学报(工学版),2010,44(7):1406-1413.
[7] 纪昌明,周婷,向腾飞,等. 基于网格搜索和交叉验证的支持向量机在梯级水电系统隐随机调度中的应用[J]. 电力自动设备,2014,34(3):125-131.
[8] 周华平,熊博杰,桂海霞. 最小二乘支持向量机优化模型在煤矿安全预测中的应用[J]. 测绘科学,2014,39(7):150-154.
[9] 张炎欣,王伟,张航. 基于即时学习策略的火电厂球磨机负荷软测量[J]. 计算机工程与应用,2012,48(7):224-230.
[10] 方帅,夏秀山,曹洋,等. 基于噪声水平估计的图像盲去噪[J]. 模式识别与人工智能,2015,28(1):51-57.
[11] 李翠平,郑瑶瑕,张佳,等. 基于遗传算法优化的支持向量机品位插值模型[J]. 北京科技大学学报,2013,35(3):837.
[12] 崔晓祥,李娟. 基于支持向量机回归的电力系统负荷建模[J]. 江苏电机工程,2012,31(3):37.
[13] ITO K, NAKAMO R. Optimiznig support vector regression hyperparameters based on crossvalidation[C]//Proceedings of the international Jonit Conference on Neural Networks,2003,3(7):2077-2081.
