一类间歇过程基于数据驱动的过程监控方法研究

发布时间：2018-05-12 19:02

本文选题：κ近邻 + 间歇过程　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：间歇过程作为现代工业生产的一种重要方式,已经被广泛地应用于染料、食品、制药等众多领域。目前,将过程监控技术应用到间歇过程中,建立监测系统对间歇过程进行异常检测,已成为工业控制领域的一个重要研究方向。本文基于κ近邻方法,开展有关间歇过程实时监测的研究,具体的研究内容如下:(1)提出一种基于κ近邻的实时监测间歇过程方法。该方法首先将间歇过程的历史数据沿着时间轴切割成多个时间片数据;然后,利用κ近邻方法基于各个时间片数据建立对应的单时刻模型,并基于这些模型实时地监测间歇过程。(2)提出一种提高过程监控实时性的快速κ近邻方法。基于E2LSH加速κ近邻算法的近邻搜索过程,利用E2LSH算法从训练数据集中剔除与当前查询样本差异较大的样本,在剩余样本中建立κ近邻监控模型进行异常检测。(3)提出一种适合过程多时段特性的κ近邻实时监测方法。首先,基于随机投影和K-均值聚类提出一种时段划分方法,该方法基于随机投影后的测量数据,利用K-均值聚类算法将间歇过程的整个生产周期划分成多个子时段;然后,利用κ近邻方法建立子时段监控模型实时地监测间歇过程。
[Abstract]:As an important way of modern industrial production, batch process has been widely used in many fields, such as dyes, food, pharmacy and so on. At present, it has become an important research direction in the field of industrial control to apply process monitoring technology to batch process and to establish monitoring system to detect the anomaly of batch process. In this paper, the real-time monitoring of batch processes is carried out based on 魏 nearest neighbor method. The specific research contents are as follows: 1) A real-time monitoring method for batch processes based on 魏 nearest neighbor is proposed. In this method, the historical data of the batch process are first cut along the time axis into a plurality of time slice data, and then the corresponding single time model is established based on each time slice data by using the 魏 nearest neighbor method. Based on these models, a fast 魏 nearest neighbor method is proposed to improve the real-time performance of process monitoring. Based on E2LSH to speed up the nearest neighbor search process of 魏 nearest neighbor algorithm, the E2LSH algorithm is used to remove samples from the training dataset that are quite different from the current query samples. In this paper, a 魏 nearest neighbor monitoring model is established for anomaly detection in the remaining samples. (3) A real-time 魏 nearest neighbor monitoring method is proposed, which is suitable for multi-time process characteristics. Firstly, a time division method based on random projection and K-means clustering is proposed. Based on the measured data after random projection, the whole production cycle of batch process is divided into several sub-periods by using K-means clustering algorithm. A subperiod monitoring model is established by using 魏 -nearest neighbor method to monitor batch processes in real time.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP274

【参考文献】