Optimized data acquisition by time series clustering in OPC

How to optimize OPC Group Management is the most frequently asked question when integrating OPC with a SCADA system. Group management assumes that the OPC client has the information to partition OPC items into homogeneous OPC groups with optimal configuration parameters, such as update rate or deadband. In reality, supervised group management mandates an empirical configuration which often leads to high group polling rate on the server and low item update rate on the client. In this paper we propose an unsupervised OPC group management concept and algorithm by modeling the OPC items as time series functions in order to quantify the similarities. Partitioning items into the optimal OPC groups is achieved using the hierarchical clustering that does not require the number of optimal clusters to be known in advance as oppose to K-mean which often produces suboptimal result and reduce the homogeneity within the group. An evaluation comparison is provided for the unsupervised and supervised method that suggests that our approach produced outstanding performance.