为实现不完备多变量时间序列的有效重构,将经典重构技术和粗糙集约简理论相结合,提出了一种广义输入状态重构方法和LS_SVM预测模型。首先,结合MeanCompleter补齐算法和经典相空间重构方法,对不完备多变量时间序列进行补齐和含有一定嵌入裕量的初始重构,以克服序列中可能存在的数据缺失和嵌入不足等问题;然后,通过构建时间序列决策表,采用一种IGA算法对冗余嵌入和冗余变量进行RS约简,获取精简重构样本空间;最后,将精简结果作为LS_SVM的输入,辨识关键变量预测模型。将提出的方法应用氧化铝配料过程的原料组份时间序列的重构和预测,通过比较和分析验证了算法的有效性和优越性。
In order to effectively reconstruct the phase space for multivariate time series, a general input state reconstruction method combining classical reconstruction technology with reduction theory of rough sets and LS_SVM prediction model are proposed. Firstly, by using Mean Completer Algorithm and classical reconstruction method, the multivariate time series with incomplete information is completed and originally reconstructed with redundant embedding dimensions. Then, the original decision-table for multivariate time series is set up and the RS reduction is implemented by IGA to delete the redundant dimensions and irrelevant variables and to extract the simplified samples. Finally, the prediction model for the key variable is identified by inputting these samples into LS_SVM. The proposed method is applied to reconstruct and predict the time series of material component contents in blending process of alumina production. The application results show its efficiency and superiority in comparison with classical method.