多向主元分析(MPCA)是监测间歇生产过程故障的较为有效的方法,但由于其自身的线性化特点,故在复杂的非线性动态系统的监测中便不能及时、准确地发现故障。本文就MPCA法的这一缺点提出新的多段MPCA法,根据过程本身的动态特性,将间歇过程分成多阶段,用多个MPCA模型来描述。此法应用于监测青霉素发酵的过程,比普通MPCA能更及时、准确地检测到故障。 Multi-direction principal component analysis (MPCA) is a more effective method to monitor the intermittent production process failure. However, due to its own linearization characteristics, MPCA can not detect faults timely and accurately in the monitoring of complex nonlinear dynamic systems. In this paper, a new multi-stage MPCA method is proposed based on this shortcoming of MPCA method. According to the dynamic characteristics of the process itself, the batch process is divided into multiple stages and described by multiple MPCA models. This method is used to monitor penicillin fermentation process, more timely and accurate than normal MPCA to detect the failure.