氧化铝蒸发浓度的自适应加权LSSVR预测  

Alumina Evaporation Concentration Prediction Based on Adaptive Weighted LS-SVR

作　　者：阳春华[1] 聂晓凯[1] 柴琴琴[1] 桂卫华[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《控制工程》2012年第2期187-190,共4页Control Engineering of China

基　　金：国家自然科学基金项目(60874069);国家863项目(2009AA04Z124)

摘　　要：针对氧化铝蒸发过程的工业现场出口料液浓度在线检测困难、操作参数具有时变性以及传统离线预测所存在的不足等特点,提出了一种多输入多输出系统的自适应加权最小二乘支持向量回归,并用于氧化铝蒸发过程出口料液浓度的在线预测。该方法根据模型预测效果自适应在线调整建模的训练样本集,利用主元分析提取主元作为分段加权支持向量回归模型的输入,采用网格搜索和交叉验证法对多输入多输出模型参数进行优化。采用工业现场的实测数据进行实验分析,计算结果表明:该方法能够很好地在线预测氧化铝蒸发过程出口料液浓度,相比基于最小二乘支持向量回归以及基于BP神经网络的浓度预测模型,该方法具有更高的预测精度和更好的泛化性能,满足实际工业生产在线优化控制要求。Considering the difficulty of online-measurement in industrial field, the time variation of operation parameters and the defi- ciency of the traditional off-line prediction in alumina evaporation process, a meihod based on adaptive weighted least square support vector regression （AWLS-SVR） is proposed for predicting the output concentration of alumina evaporation process in multiple-input mul- tiple-output （MIMO） system. Training data set required for modeling is adjusted online adaptively according to the prediction results. The principle components extracted by principle component analysis （PCA） is taken as the input of the segmented weighted support vec- tor regression model. The parameters of the MIMO model are optimized by grid search and cross validation method. Simulation results based on the data from industrial field show that it can better predict online the output concentration of alumina evaporation process. Compared with the LS-SVR model and BPNN model, ~ has higher accuracy and better generalization, and it can satisfy the require- ments of online optimal control in practical industrial production.

关 键 词：氧化铝蒸发过程 多输入多输出 最小二乘支持向量回归 在线预测 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]