APP下载

基于改进的高斯混合回归的球磨机料位软测量

2018-03-03杨飞乔铁柱庞宇松阎高伟

现代电子技术 2018年5期

杨飞+乔铁柱+庞宇松+阎高伟

摘 要: 针对球磨机系统多模态复杂过程中的料位不确定性,球磨机振动信号存在非线性、噪声和外界干扰等问题,采用一种基于改进的高斯混合回归(GMR)的球磨机料位软测量方法,解决传统高斯混合模型初始化含有噪声和异常值的数据难以聚类的问题。首先,利用改进的K?medoids聚类算法与EM算法分别初始化和优化高斯混合模型(GMM)的最佳高斯分量个数、最优模型参数,然后采用GMR预测输出球磨机料位。最后实验验证了改进GMR模型得到的预测料位可以很好地跟踪真实料位,并且通过实验结果的对比分析,验证了改进模型的有效性和实用性以及较好的预测精度。

关键词: 球磨机料位; 多模态; 振动信号; GMM; 聚类; 软测量; GMR

中图分类号: TN98?34; TP181 文献标识码: A 文章编号: 1004?373X(2018)05?0153?06

Abstract: Since the fill level of the ball mill system in multimode complicated process is uncertain, and the vibration signal of ball mill has the characteristics of nonlinearity, noise and outside interference, a soft measurement method for ball mill fill level based on improved Gaussian mixture regression (GMR) is proposed to solve the problem that it is difficult to cluster the data embedding noise and abnormal value of the traditional Gaussian mixture model (GMM) initialization. The improved K?medoids clustering algorithm and EM algorithm are used respectively to initialize and optimize the optimal Gaussian component quantity and optimal model parameters. The GMR is used to predict the output level of the ball mill. The experimental results verify that the predicted fill level obtained by improved GMR model can track the real fill level accurately. The comparative analysis of experimental results verifies that the improved model is feasible and practical, and has high prediction accuracy.

Keywords: ball mill fill level; multimode; vibration signal; Gaussian mixture model; clustering; soft measurement; Gaussian mixture regression

0 引 言

球磨機是磨矿工业中广泛使用的基础设备,准确测量球磨机料位是实现优化控制和节能降耗的关键技术之一。近年来,科研人员提出了许多测量球磨机料位的方法,其一般步骤是:先对采集的信号计算频谱特征,然后进行降维和特征提取,最后建立频谱特征与料位之间的模型。文献[1]利用快速傅里叶变换方法求取信号的功率谱密度(Power Spectrum Density,PSD),然后采用主元分析(Principal Component Analysis,PCA)方法提取特征和减少冗余信息,最后用支持向量回归(Support Vector Regression,SVR)建立回归模型。文献[2]利用PCA方法进行振动信号的特征提取,然后采用偏最小二乘回归方法(Partial Least Square Regression,PLSR)建立球磨机料位软测量模型。文献[3]采用偏最小二乘法(Partial Least Square,PLS)对球磨机振动信号进行特征提取和降维,然后采用极限学习机(Extreme Learning Machine,ELM)建立料位软测量模型。……

登录APP查看全文