针对氧化铝配料过程中返回物料成分波动大且难以在线检测的问题,首先,利用滞后的离线分析获得的多变量时间序列,直接构造包含充分预测信息的初始相空间;然后,构建时间序列决策表,并采用一种IGA算法对冗余嵌入和冗余变量进行RS约简,获取广义重构相空间;最后,根据广义重构结果构造输入样本集,建立LS_SVM实时预测模型。仿真结果表明,提出的模型具有较好的泛化能力,能获得较理想的返料成分含量预测精度(6种氧化物的相对均方根误差均小于13%),具有一定的应用价值。
Considering the great fluctuation and the difficulty of on-line measurement of oxide contents of returned materials in the blending process of alumina production, the original reconstruction phase space containing enough information for prediction was firstly constructed by directly using multivariate time series from the off-line analysis. Then, the original decision-table for multivariate time series was set up and the RS reduction was implemented by IGA to delete the redundant dimensions and irrelevant variables and to reconstruct the generic phase space. Finally, the input samples were extracted according to generic reconstruction results and LS_SVM was built to real-time predict oxide contents of returned materials. Simulation results show that the developed model has better generalization ability and satisfactory prediction accuracy (the relative root mean square errors of six oxides are lower than 13%), so it is feasible and worthwhile for application.