| 英文摘要 | Dynamic principal component analysis(DPCA) extracts the time-serial auto-correlation inherited from sampled data through enhancement matrix or vector. However, the way of monitoring the auto-correlated features and residuals directly in the DPCA model is not appropriate, given that the negative influence caused by the auto-correlation on the monitoring statistics is ignored. Therefore, on the basis of the DPCA model that exacts time-serial auto-correlation, how to eliminate the auto-correlation inherited in the sampled data is further required. This paper proposed a dynamic process monitoring method based on estimated errors. Through sequentially assuming the measured data of each process variable was missing, it introduced and incorporated the iteration method(IM) with the built DPCA model so as to calculate the estimates of corresponding variable. Since the estimates could approximate the original measured data to a large extent with only one variable was missing, the inconsistency between the two(i. e, estimation error) no longer existed obvious auto-correlation. Moreover, the variation of the estimation error could directly reflect the abnormal variation in the sampled data, the estimation error could thus be used for dynamic process monitoring purposes. Finally, through comparisons in two dynamic examples, a dynamic numerical process and the Tennessee Eastman benchmark process, the superiority of the proposed DPCA-IM approach over other dynamic process monitoring methods are validated. |