Grid Stabilization With Distributed Assets
The integration of renewables, the decentralization of generation, and the growing number of power electronics devices in the grid are turning power systems into highly dynamic, complex environments. In this new paradigm, ensuring frequency stability after a disturbance is no longer a trivial task.
Background
The job of Transmission System Operators has been to gather many signals, process them, and output the right control actions in real time. However, in a distributed environment, how are you supposed to maintain balance in the grid when there are thousands of intelligent devices that can make decisions on their own?
Now imagine an ideal scenario where every asset in the grid, including storage systems or converter-interfaced renewables with grid-following/grid-forming controls, to name a few, react locally, based on their own capabilities, but in perfect coordination with the rest of the system. No central controller, just collective intelligence.
That is what we are building with COLMENA, a project developed by the Barcelona Supercomputing Center in collaboration with eRoots. In this context, COLMENA introduces a fully distributed control system based on multiple agents that allows the grid to regulate itself in real time, intelligently, autonomously, and robustly.
Model Predictive Control
The core of COLMENA is a network of software agents, each linked to a controllable device. Every agent is capable of executing different roles based on measurements and internal calculation rules rules, making them capable of selecting control actions that benefit both their local asset and the overall grid. Such is the case with the frequency, a magnitude that can be locally measured but is inherently global. This is achieved by applying a distributed Model Predictive Control (MPC).
Distributed MPC enables each agent to:
- Predict how the grid frequency will evolve.
- Coordinate with neighboring agent.
- Make control decisions that contribute to the global frequency stability. These control decisions include:
- Changing role of converter, between grid following and grid forming
- Load Shedding
- Activating the frequency response of generators
The method works in a loop:
- Each agent collects local measurements and shared information.
- They independently solve a local optimization.
- Through iterative coordination rounds, they converge to an agreement on their control actions.
- These actions are implemented in real time, and the cycle repeats.
This architecture requires no central coordinator and is naturally scalable. Even under severe disturbances, such as generator trips, or short-circuit faults, the system can reorganize control decisions on the fly and maintain frequency close to its nominal value.
Results and next steps
We have tested COLMENA in large-scale simulations with promising results. Under various contingency scenarios, frequency deviations were quickly suppressed and control efforts were smoothly distributed across all assets.
Ideally, the next step is the transition from simulation to real-world demonstration, perhaps integrating hardware-in-the-loop testing and preparing COLMENA for deployment in microgrids, and eventually, utility-scale systems. Also, the actual MPC we have implemented relies on the DC approximation of the network. As we always preach, including non-linearities is eventually the way to go.
At eRoots, we believe the grid of the future will not be centrally controlled — it will be collectively controlled. With the vision of the Barcelona Supercoputing Center, COLMENA is a step toward making that vision real.
