自回归求和滑动平均方法用于间歇过程变量在线预测  

Online Prediction of Batch Process Variables Using Data Reconstruction-Autoregressive Integrated Moving Average Method

作　　者：刘兴红[1] 邹志云[1] 刘景全[1] 郭宇晴[1] 于鲁平[1] 

机构地区：[1]防化研究院,北京102205

出　　处：《石油化工自动化》2012年第2期41-44,共4页Automation in Petro-chemical Industry

摘　　要：间歇过程变量的在线预测是一种重要的生产过程质量控制手段。实现间歇过程变量的在线预报需要对过程以往的批次数据建立预测模型,即需挖掘批次间和批次内的数据信息。针对间歇过程数据不同批次不等长、数据长度短、非线性等特点,采用数据重构——自回归求和滑动平均方法建立其在线预测模型:将收集到的间歇过程变量以批次为单位进行数据平滑;对这些批次数据按照随机的顺序首尾相接,组成长数据集;对于批次连接处数据跳跃的情况,采用后面所有批次数据减去上一批次的最后一个值,以实现数据的平滑;采用自回归求和滑动平均方法建立数据模型,并用于间歇蒸馏温度的在线预报。采用该方法建立的4步预测模型对某间歇蒸馏过程上升气温度的预测均方差较小,符合生产现场的预测要求。Online prediction of batch process variables is an important quality control means for production process. The production data with previous batches are in need to develop the prediction model for realizing online prediction of process variables, that is to mine the relationship among different batches and within the batches. A nonlinear online prediction model based on data reconstruction-autoregressive integrated moving average method is proposed according to the batch process data characters of nonlinearity, short data length and unequal data lengths of different batches. The collected batch process data are smoothed as batch unit. All these batch data are linked to each other randomly from the very beginning to the end, and form a long data collection. The uncontinuity of two continuous hatches is processed by subtracting the last data of previous batch to realize the data smooth. A data model is developed using autoregressive integrated moving average method, and used for the online prediction of batch distillation temperature. The mean square error with the developed four step prediction model for the updraft temperature of one batch distillation process is proved to be precise and exact, and consistent to the prediction requirement of the site.

关 键 词：间歇过程 自回归求和滑动平均 非线性时间序列预测 

分 类 号：O211.61[理学—概率论与数理统计] O213.1[理学—数学]