数据挖掘技术用于间歇蒸馏过程转馏分点的在线预报

Online prediction of the distillate transferring point in batch distillation process using data mining technology

对于间歇蒸馏过程,提前准确判断从低馏分到主馏分的转馏分点是影响最终产品质量和产量的关键环节,设计基于数据挖掘技术的转馏分点在线预报软测量系统是一项重要的过程质量控制手段,为提高生产的综合自动化水平创造了重要条件。根据混沌理论,温度能较高程度的反映体系内反应及分离情况,因此选取间歇蒸馏上升气温度为考察变量。针对数据非线性、动态、数据长度短、不同批次数据不等长等特点,提出了将不同批次数据按照随机的顺序首尾相接组成长数据集的数据重构策略;采用自回归求和滑动平均方法和最小二乘支持向量机方法建立了组合时间序列预测模型;通过对理论转馏分温度与实际转馏分温度的差值和预测曲线近似斜率的统计分析,建立了转馏分点在线预报系统,经过在实际生产中的验证,实现了对转馏分点提前1min的准确预报。

Estimating the distillate transferring point from light mixture to main product is very important for batch distillation process. Soft measurement system for the distillate transferring point is an important process quality control strategy, which could increase the production automation level. The updraft temperature of the batch distillation process is chosen to be study variable, because according to chaotic theory, temperature can reflect on generally the reaction and separation situation. A data reconstruction method which can be used to preprocess the nonlinear and dynamic batch process variables, whose length are short and unequal, is proposed. This method is that process data of all batches is randomly linked to each other to build a long data series. Autoregressive integrated moving average-least square support vector machine hybrid nonlinear time series prediction model is developed, and the mean square error from this model is smaller than from the autoregressive integrated moving average model. The difference of theoretical and real distillate transferring point, and the approximate slope of prediction curve is statistically analyzed, and then a prediction system for the distillate transferring point is established. The developed prediction system is used in real production process, and is proved to to be precise and exact.