摘要
针对传统球磨机料位软测量模型难以适应生产过程中的时变特性,以及磨机信号中存在的非线性和不确定性等问题,将磨机筒体的振动和振声频谱作为辅助变量,提出基于改进即时学习算法的料位软测量模型。采用梅尔频率倒谱系数提取所有样本的特征参数;以当前样本作为查询样本,采用云模型理论计算其与历史样本的相似度并得到最近邻样本;利用得到的最近邻样本建立局部最小二乘支持向量机模型并预测当前料位。实验结果表明,该软测量模型可以有效地实时测量球磨机料位并能获得较高的预测精度。
The traditional soft sensor models of ball mill fill level are difficult to adapt to the time-varying characteristics in the industrial process.Moreover,the nonlinear and uncertainty universally exist in ball mill signals.Aiming at these problems,a soft sensor model based on enhanced just-in-time learning was proposed for fill level prediction,in which vibration and acoustic spectrum of the drum were regarded as the auxiliary variables.Mel-frequency cepstrum coefficient was employed to extract features from all samples.The nearest neighbor dataset of current sample was selected from the historical samples by calculating similarity measurement of cloud model.The least square support vector machine for locally modeling was built based the nearest neighbor dataset and the current level was predicted.Experimental results show that the proposed soft sensor model can effectively mea-sure the fill level of ball mill and obtain higher prediction accuracy.
作者
贾松达
丁洁
阎高伟
JIA Song-da;DING Jie;YAN Gao-wei(College of Information Engineering,Taiyuan University of Technology,Taiyuan 030024,China)
出处
《计算机工程与设计》
北大核心
2018年第4期1011-1016,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61450011)
山西省自然科学基金项目(2015011052)
山西省煤基重点科技攻关基金项目(MD2014-07)
关键词
即时学习
软测量
不确定性
云模型相似性度量
最小二乘支持向量机
just-in-time learning
soft sensor
uncertainty
similarity measurement of cloud model
least square support vector machine
