基于k近邻的批次过程在线实时监测方法

针对基于k近邻的故障检测方法(Fault Detection method using the k-Nearest Neighbor rule,FD-kNN)的在线实时监测需预估当前时刻之后的采样数据,检测性能会受到预估精度影响的问题,对FD-kNN进行扩展以适用于批次过程的实时监测.该方法根据每个采样时刻的历史数据进行建模,并根据这些模型实时监测批次过程.该方法不需要预估数据,避免由于预估误差大而带来的误报和漏报问题,同时较好地继承k近邻法则(k-Nearest Neighbor rule,kNN)在处理非线性、多模态和非高斯等问题上具有的优势.青霉素发酵过程的仿真试验验证该方法可行.

Online real-time monitoring method based on k-nearest neighbor for batch processes

In recent years, fault detection using the k-Nearest Neighbor rule (FD-kNN) has been proposed and applied to batch processes successfully. The FD-kNN adopts the way of off-line monitoring, however, if it is applied to online real-time monitoring, it has to estimate the measurement data from the next sampling time to the end of the batch. Therefore, the fault detection performance will be significantly affected by the estimated accuracy. To overcome this problem, this paper extends the FD-kNN method in order to apply to the real-time monitoring for batch processes. The proposed method builds a model at each sampling time based on the data at each time slice and monitors the batch process in real time based on these models. It does not need to estimate the measurement data, and avoids false alarms and missing detections due to the large error which may occur in estimating the data. Moreover, the proposed method inherits the advantages of kNN in dealing with the nonlinear, multimode and non-Gaussian problems. Finally, the experiment on penicillin fermentation process illustrates the effectiveness of the proposed method.