基于CART-LSSVM的球磨机料位软测量方法研究

球磨机是用于电力、磨矿和冶金等行业的高能耗设备,准确测量其滚筒料位能够提高运行效率和安全性能.针对其滚筒内料位难以实时检测,球磨机的轴承振动信号中存在较多的冗余特征,提出了一种基于分类回归树和最小二乘支持向量机的软测量方法,首先用Welch法获得振动信号的功率谱密度,并分割得到基本特征,然后建立分类回归树模型,根据最优树模型的分支节点进行特征选择,最后利用最小二乘支持向量机实现特征变量与料位间的非线性映射.通过实验结果的对比分析,验证了该模型的有效性和实用性,以及良好的预测精度.

Soft Sensor for Ball Mill Fill Level based on CART-LSSVM Model

Ball mill is a high energy consumption equipment used in electricity,grinding and metallurgical indus-tries.Accurate measurement of its fill level(FL)can improve operational efficiency and safety performance. Howev-er,The real-time measurement of FL is difficult to realize,and the components of bearing vibration of ball mill are complex and redundant. Aiming at these problems,a new soft sensor approach of FL based on Classification and Re-gression Tree(CART)and Least Squares Support Vector Machine(LSSVM)is proposed. Firstly,the Power Spectrum density(PSD)of bearing vibration is obtained by welch method,essential features are achieved by partition subse-quently. Secondly,these features are adopted to build CART,and branch nodes of the best model is selected as fea-tures. Finally,the LSSVM are used to implement the non-linear mapping between features and FL. The comparative experiments verifies that this model is feasible and practical with high prediction accuracy.