空分过程变量预测的多变量时间序列分析方法

Multivariate time series based variable prediction in air separation process

氮塞是空分过程的常见故障,粗氩塔冷凝器出口氩气含氩量是工业现场中指示氮塞是否发生的关键变量,对该变量进行准确的预测可以使氮塞故障的报警时间提前。本文采用多变量时间序列相空间重构的方法,建立了粗氩塔冷凝器出口氩气含氩量和其它过程变量之间的一步线性回归预测模型,以迭代方式获得多步预测的结果,并利用滑动窗口实现了模型参数的在线修正。通过某钢铁公司空分装置实际数据的建模与仿真,分析了相空间重构时嵌入维数以及预测步数的选取对最终预测结果的影响,即预测均方误差与嵌入维数成反比,与预测步数成正比。仿真结果同时表明,本文建立的模型能够较为准确地对空分过程关键变量进行预测,预测提前时间在4～5分钟之间。

Nitrogen blockage is a common fault in air separation process.The molecular volume component of argon in the outlet of the condenser in the crude argon column is often used as an indicator of nitrogen blockage.Accurate prediction of this variable is of great help to predict the occurrence of nitrogen blockage in advance.A predictive model based on the phase-space reconstruction of multivariate time series is presented in this paper.The one-step ahead predictive model is established by the least squares method and the multi-step ahead prediction results of the process variable are achieved iteratively via one-step ahead prediction.Moreover,an on-line updating of the model parameters is implemented by a moving window.The embedding dimension of the phase-space reconstruction and the steps ahead are analyzed based on the practical data from an industrial air separation process.Analysis shows that the mean square prediction error is inversely proportional to the embedding dimension of the phase-space reconstruction while derectly to the steps ahead.The simulation results also demonstrate that the proposed model can forecast the predicted process variable accurately,the predicted time could be between 4 and 5 minutes.