Power grid probing for load learning: Identifiability over multiple time instances

Although nodal loads and renewable generation need to be known for solving different grid optimization tasks, power distribution networks currently lack extensive metering infrastructure. The fresh idea here is to exploit the capabilities of smart inverters found in solar panels and energy storage devices to probe the grid and thus infer the power injections at non-metered nodes. Probing can be accomplished by commanding inverters to change their active injections and power factors momentarily, and subsequently record the incurred voltages as responses of the underlying physical system. Load inference via grid probing boils down to an implicit nonlinear system identification task and involves solving a sequence of power flow problems coupled over multiple probing periods. Invoking results from structured control, an intuitive rule pertaining to the locations of metered and non-metered nodes on the grid graph is shown to be sufficient for ensuring local identifiability in radial and meshed networks. The identifiability criterion scales favorably over time and is numerically corroborated on the IEEE 34-bus feeder.