基于维度压缩和聚类分析的化工报警阈值优化研究

发布时间：2018-04-22 16:28

本文选题：报警阈值 + 优化　；参考：《青岛科技大学》2017年硕士论文

【摘要】：现代化工生产中,为了提高生产过程的安全性和稳定性,通常需要使用报警管理系统对一些过程变量进行报警阈值设置。阈值设置不合理会产生过多无效报警,增加操作负荷,严重时会引发事故,导致报警系统失效。因此,对设置不合适的变量报警阈值进行优化是十分必要的。本文通过TE(Tennessee Eastman,田纳西-伊斯曼)过程和某工业原油常减压操作实例,对多变量报警阈值优化新方法进行了研究。针对多变量报警阈值优化方法,本文主要做了两方面研究。一方面,提出了基于PCA(Principal Component Analysis,主成分分析)权重和Johnson转换的多变量报警阈值优化方法。通过PCA计算变量权重,对变量数据进行Johnson正态转换,利用概率密度估计求出FAR(False Alarm Rate,误报率)和MAR(Missed Alarm Rate,漏报率),在满足FAR降低且报警数目不超过国际标准中规定的单位时间内限制的报警数目(通常平均每分钟不超过一个报警)的情况下优化报警阈值。另一方面,提出了基于报警聚类和ACO(Ant Colony Optimization,蚁群优化)的多变量报警阈值优化方法。通过标准化欧式距离(Euclidean Distance,欧几里得距离)实现报警聚类,利用熵权法求出变量权重,拟合出变量在正、异常状态下的概率密度函数。添加报警延时,建立关于误报率、漏报率和AAD(Average Alarm Delay,平均报警延时)的目标函数,利用ACO算法优化目标函数。这两种方法在一定程度上都实现了优化阈值的目的。通过TE过程和某工业原油常减压操作实例对本文研究方法进行了验证。结果表明,与传统方法相比,该方法更能有效减少报警次数和报警率,在报警阈值优化方面具有优势。
[Abstract]:In modern chemical production, in order to improve the safety and stability of the production process, alarm management system is usually used to set the alarm threshold for some process variables. The unreasonable setting of threshold will cause too many invalid alarms, increase the operating load, and lead to accidents when serious, which will lead to the failure of alarm system. Therefore, it is necessary to optimize the setting of inappropriate variable alarm threshold. Based on the TE(Tennessee Eastman (Tennessee Eastman) process and an example of an industrial crude oil operating under atmospheric and vacuum pressure, a new method of multivariable alarm threshold optimization is studied in this paper. Aiming at the optimization method of multivariable alarm threshold, this paper mainly researches on two aspects. On the one hand, a multivariable alarm threshold optimization method based on PCA(Principal Component Analysis (PCA) weight and Johnson conversion is proposed. The variable weight is calculated by PCA, and the variable data is transformed by Johnson normality. Using probability density estimation to calculate FAR(False Alarm rate (false alarm rate) and MAR(Missed Alarm rate, false alarm rate, the number of alerts (usually not exceeding the average per minute per minute) within the limit of the number of alerts per unit time specified in the international standard for satisfying the decrease of FAR and not exceeding the limit per unit time specified in international standards An alarm) is optimized in the case of an alarm threshold. On the other hand, a multivariable alarm threshold optimization method based on alarm clustering and ACO(Ant Colony optimization (ant colony optimization) is proposed. The alarm clustering is realized by Euclidean distance (Euclidean distance), the weight of variables is calculated by entropy weight method, and the probability density function of variables in positive and abnormal state is fitted. The objective function of false alarm rate, false alarm rate and AAD(Average Alarm delay (average alarm delay) is established by adding alarm delay. ACO algorithm is used to optimize the objective function. To some extent, these two methods achieve the purpose of optimizing the threshold. The method of this paper is verified by te process and an example of an industrial crude oil operating at atmospheric and vacuum pressure. The results show that compared with the traditional method, this method can effectively reduce the alarm frequency and alarm rate, and has advantages in the optimization of alarm threshold.
【学位授予单位】：青岛科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TQ086

【参考文献】