基于Unscented卡尔曼滤波新息的多变量序贯概率比检验故障检测方法

Fault detection method based on multi-variable sequential probability ratio test of Unscented Kalman filter innovation

针对非线性多变量过程监控问题,提出一种基于预测新息的多变量序贯概率比检验(SPRT)方法。首先利用Unscented卡尔曼滤波(UKF)基于正常过程模型预测输出值,将预测输出值与过程测量得到的实际输出值对比产生预测新息,然后引入多变量SPRT方法分析多元新息的统计特性,构造对数概率似然比判决函数和判决规则,监控过程的运行状态并对故障状态进行报警。在连续搅拌反应器上的仿真应用结果表明,所提出的故障检测方法能够有效实现过程监控,比传统的残差加权平方和方法误报率低、检测速度快。

A multi-variable sequential probability ratio test（SPRT）method based on predictive innovation was proposed for nonlinear multi-variable process monitoring problem.Firstly Unscented Kalman filter（UKF） is conducted to predict outputs using normal process model,the predictive innovation is generated by comparing predictive outputs and the actual ones which are sensed from the process.Then multi-variable SPRT method is introduced to analyze the statistical characteristics of the multi-dimension innovation.Decision function and decision rules with log-probability likelihood ratio are constructed to monitor the status of the process and signal the faults.The simulation results on continuous stirred tank reactor show that the proposed method can monitor process effectively.Compared with the traditional weighted-sum squared residual method,the proposed method has low false alarm rate and detects faults quickly.