Robustness and Safety of Smart-Grid Real-Time Control

Contact: Andrey Bernstein

We consider a distributed real-time control of the smart power grid.
Each part of the grid is assigned to a control agent whose goal is to optimize the performance and the quality of service of the grid by continuously controlling the devices under its responsibility.
In order to do so, the agent receives messages that contain the data required for this control, such as measurement, feasible areas of operation, etc.

Such a system is envisioned to be deployed at the EPFL Smart-Grid in the near future.

One of the main challenges in this setting is to ensure normal operation even in face of the failure of some of the components.
Traditionally, this is done by introducing replicated copies of the same component which perform the same task, but are implemented using different hardware, software, and by different development teams.
This is how the crucial systems are implemented, for instance, in the airplane control system.

For simplicity, consider a controller that receives data from three replicated components.
Ideally, this data should be identical. But in practice it may differ due to: (i) numerical precision error, and (ii) bugs in the implementation of a replicated component.

The goal of this project is to explore the different ways to combine the data received from the replicated components.
In particular, the data will be represented as a region of feasible operation in two dimensions (that correspond to the active and reactive power in the grid).
Given three sets, received from the three replicated components, the goal is to correctly choose the best set among them.

The methods should be robust to one outlier: if one of the three sets is “completely different”, it should not be chosen in any case.
The difference between the sets will be measured using different set-to-set distance metrics, and the corresponding methods will be evaluated numerically.
Moreover, since we are interested in real-time control, we will seek for methods with low computational complexity.

Required skills:

Linear algebra, Programming, Simulation