We design a class of ad-hoc data fusion algorithms that can exploit and extract reliable values from heterogeneous measurement data streams. These algorithms retrieve and identify reliable true values in an ad- hoc manner whenever required, to ensure provable time cost and response time for real-time learning. (2) We devise a class of online stream learning algorithms to estimate virtual measures. These algorithms seamlessly integrate data-driven and physical models to fine-tune model parameters, and perform online model updates only when necessary, to ensure the feasibility over fast measurement data streams. (3) The proposed Virtual Meter closes the loop of interaction between data-driven methods and physics-based methods to provide enhanced estimation of real-time value of loads and states of power distribution systems.
» More ...