| 引用本文: | 汤健,柴天佑,赵立杰,岳恒,郑秀萍.基于振动频谱的磨矿过程球磨机负荷参数集成建模方法[J].控制理论与应用,2012,29(2):183~191.[点击复制] |
| TANG Jian,CHAI Tian-you,ZHAO Li-jie,YUE Heng,ZHENG Xiu-ping.Ensemble modeling for parameters of ball-mill load in grinding process based on frequency spectrum of shell vibration[J].Control Theory and Technology,2012,29(2):183~191.[点击复制] |
|
| 基于振动频谱的磨矿过程球磨机负荷参数集成建模方法 |
| Ensemble modeling for parameters of ball-mill load in grinding process based on frequency spectrum of shell vibration |
| 摘要点击 2639 全文点击 2457 投稿时间：2011-05-04 修订日期：2011-09-21 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/j.issn.1000-8152.2012.2.PCTA110478 |
| 2012,29(2):183-191 |
| 中文关键词 磨矿过程 球磨机负荷 频谱聚类 核偏最小二乘 信息熵 集成建模 |
| 英文关键词 grinding process ball-mill load frequency-spectrum clustering kernel partial least squares information entropy ensemble model |
| 基金项目 国家自然科学基金资助项目(61020106003); 中国博士后自然科学基金资助项目(20100471464); 国家自然科学基金资助项目(60874057); 国家科技支撑计划资助项目(2008BAB31B03). |
|
| 中文摘要 |
| 以磨矿过程的湿式球磨机为背景, 针对传统磨机负荷(ML)检测方法只能依靠灵敏度较低的轴承振动、筒体振声和磨机功率等信号监督判断ML状态, 难以检测磨机内部负荷参数的问题, 提出了一种基于高灵敏度的筒体振动频谱的集成建模方法. 首先, 依据磨矿过程的研磨机理, 将振动频谱采用波峰聚类方法自动划分具有不同物理意义的分频段; 然后利用核偏最小二乘(KPLS)算法分别建立各分频段的ML参数子模型; 最后, 依据子模型训练数据预测误差的信息熵获得初始权重, 加权得到最终的ML参数集成预测模型; 在线使用中则根据子模型预测误差的变化进行权值的在线自适应更新. 仿真结果证明了该方法的有效性. |
| 英文摘要 |
| The mill load (ML) of a wet ball-mill is generally determined by using the low-sensitivity shaft vibration signal, acoustical signal from mill shell and mill power during the grinding process. The result doesn’t reflect the status of the load parameters inside the ball-mill. To deal with this problem, we propose the ensemble modeling method based on the high-sensitivity frequency-spectrum signal from the shell vibration. According to the grinding mechanism, the vibration frequency spectrum is automatically partitioned into several different spectral segments with different physical meanings based on the frequency-spectrum clustering method. The sub-model of ML parameters in each spectral segment is built by using the kernel partial least-squares (KPLS) algorithm. The final ML parameters ensemble prediction model is obtained as the weighed combination of sub-models. Each initial weighting coefficient is determined from the information entropy of the training data prediction error of the corresponding sub-model. The weighting coefficient can be adaptively updated online based on the variation of prediction error of the sub-model. Simulation results demonstrate the validity of the proposed method. |
|
|
|
|
|