Abstract
Many real systems have a network/graph structure with many connected nodes and many edges representing deterministic or stochastic dependencies and interactions between nodes. Various types of known or unknown anomalies and disturbances may occur across these networks over time. Developing real-time anomaly detection and isolation frameworks is crucial to enable network operators to make more informed and timely decisions and take appropriate actions to improve operations, maintenance schedules, and overall safety of the system and prevent operation disruptions due to anomalies. To monitor the health of modern networks in real-time, different types of sensors and smart devices are installed across these networks that can track real-time data from a particular node or a section of a network.
In the first phase of my research, an innovative Bayesian Network (BN) based inference method is introduced to calculate the most probable explanation (MPE) of a set of hidden nodes in sensor-driven networks. Given the values of data from a set of observable sensors installed at a subset of nodes in the network and the stochastic relationships between hidden nodes and instance nodes, the estimation of the network status will be computed. Furthermore, the novel use of an effective parallel sampling method and vectorization accelerates the computation process and makes the proposed framework applicable to large-scale networks. The efficiency of the approach is shown by a comprehensive set of numerical examples.
When it comes to real-world situations, we always strive to reduce uncertainty and improve the accuracy of our predictions, whenever possible. This can be done by sending technicians or electric devices to conduct the physical inspections. The second phase of my research involves the optimization problem of sensor inspection to improve the accuracy of anomaly detection with binary sensor data through optimal sensor selection and inspection in a network. The effectiveness will be proved by a set of numerical experiments.