摘要
针对采用传统极限学习机在球磨机料位软测量建模过程中,存在鲁棒性差,预测精度不高等缺点,提出一种基于最优定界椭球(Optimal Bounding Ellipsoid,OBE)改进极限学习机(Extreme Learning Machine,ELM)的建模方法.该方法以球磨机振动信号为观测变量,采用偏最小二乘法提取有效特征,将提取到的有效特征输入到ELM中进行模型训练,并利用OBE在模型误差未知但有界的条件下,对网络权值进行约束优化.通过小型球磨机实验表明,在对球磨机料位进行回归预测时,该方法的评价指标与其它方法相比有所提高,测量结果的箱线图也直观展示该方法具有更好的鲁棒性.
In the process of soft sensor modeling of ball mill fill level using traditional extreme learning machine,existing the issue of poor robustness and low accuracy.To solve the problem,an improved extreme learning machine(ELM)soft sensor method based on optimal bounding ellipsoid(OBE)was proposed.The ball mill vibration signal was viewed as observed variables,and the features were extracted by partial least squares(PLS).Then,extracted features were put into ELM for model training.OBE was used to optimize the weights of the network under the condition that the model error was unknown but bounded.The experiment tested on a dataset of the lab-scale ball mill illustrate that the evaluation index is improved in the prediction of the ball mill fill level,and the box-plot shows that the proposed method has better robustness.
出处
《中北大学学报(自然科学版)》
北大核心
2017年第5期574-579,598,共7页
Journal of North University of China(Natural Science Edition)
基金
国家自然科学基金资助项目(61450011)
山西省煤基重点科技攻关项目(MD2014-07)
山西省自然科学基金资助项目(20150110052)
关键词
球磨机料位
软测量
最优定界椭球
极限学习机
fill level of ball mill
soft sensor
optimal bounding ellipsoid
extreme learning machine