基于Savitzky-Golay滤波器的多变量监控报警方法研究

A multivariate alarm system based on the Savitzky-Golay filter

工业报警系统是现代流程工业中集散控制系统的核心组成部分。由于不恰当的报警变量设置,实际当中常常产生大量干扰报警,对生产的安全高效形成严重威胁。造成大量干扰报警的一个重要原因是报警变量的设置缺乏与相关变量的关联。文中采用将定性趋势分析与数据驱动相结合的方法进行多变量报警系统设计。主要思想是通过Savitzky-Golay滤波器提取相关过程变量信号的滤波后幅值及各阶导数作为定性趋势的代替特征,并在特征空间中训练支持向量机模型,判别是否应当给出报警。所展示的给水泵的实例表明,文中所提出的方法具有高准确率的优点,改进了传统的对单变量设置报警阈值的方法,使得误报警、漏报警数量都大幅下降。

Industrial alarm systems are critical part of distributed control systems. Due to inappropriate settings for alarm variables, many nuisance alarms occur in practice, which do harm to safety and efficiency. A main cause to nuisance alarms is that the alarm design is isolated from related variables. A multivariate alarm system is designed by combining the qualitative trend analysis with data driven methods. The main idea is to extract the filtered signal amplitudes and deriatives of related process variables as the qualitative trend features using the Savitzky-Golay filter, and to train the support vector machine model in this feature space in order to determine rising alarms. The industrial example of a feed-water pump illustrates the accuracy of the proposed method. The false alarm rate and the missing alarm rate of proposed method are much lower than the classical univairate alarm method.