Combating Curse of Dimensionality in Resilient Plant Monitoring Systems: Overlapping Decomposition and Knowledge Fusion

摘要

Resilient plant monitoring systems (RPMS) are sensor networks that degrade gracefully under cyber-physical attacks. In the previous work, we have developed an adaptive four-layer RPMS architecture and evaluated its performance under various attack scenarios. While the steady state performance of this system has been shown to be satisfactory, the transients have not: adaptation time grows exponentially as a function of the number of states in the network. The current paper is intended to provide a method for combating this curse of dimensionality. The approach is based on the idea of overlapping plant decomposition and subsequent fusion of knowledge derived in the overlapping subnetworks. In this paper such a monitoring system is developed (five-layer architecture), analyzed, and shown to have desirable steady state and, to a certain extent, transient characteristics.