An Adaptive Sampling Strategy for Online Monitoring of Partially Observed Networks

Networks are widely used to capture interactions among different components in complex systems. It is critical to monitor them in real time to detect changes quickly and take corrective actions. For example, monitoring the power grid enables operators to address problems before they lead to outages. A challenge in network monitoring occurs when the network system can only provide information about a small subset of nodes at any given time due to resource constraints. If this is the case, at each acquisition time, a decision needs to be made on which nodes to collect data from to maximize the change detection capability. This article proposes a new adaptive sampling strategy for online monitoring of partially observed networks. First, a Gaussian process with a novel spatial kernel that exploits the global network structure is developed and combined with a temporal kernel to capture the spatio-temporal information within the network. This information is used to guide the monitoring scheme and the adaptive sampling strategy. The main idea of the adaptive sampling strategy is to balance exploration and exploitation to decide where to collect data at each acquisition time. The performance of the proposed framework is demonstrated through simulations and case studies.