Human-on-the-loop Resilient Control of InverterBased Resources Under Actuator Degradation

Imagine the electrical grid as a giant, invisible nervous system that powers our modern world, sending energy from solar panels and wind turbines to our homes and schools. For a long time, this system relied on massive, heavy machines like traditional power plants. But today, we are swapping those heavy machines for sleek, fast, and smart "inverter-based resources" (IBRs)—think of them as the nervous system's new, high-speed digital neurons. These are the devices inside your solar roof or your home battery that turn stored energy into electricity for the grid. The big challenge scientists face is what happens when these digital neurons get sick or damaged. If a solar panel's inverter starts to fail, or if a cyber-attack weakens its ability to push power, the whole neighborhood's voltage can wobble, lights might flicker, and the system could crash. To fix this, engineers usually try to build robots that automatically detect the sickness and fix it instantly. But these robots can be too rigid; they might panic and shut down, or they might try to fix a problem they don't fully understand, making things worse.

This paper introduces a clever new idea: instead of relying solely on a robot, let's put a human in the loop. The authors, Majid Dehghani, Taha Saeed Khan, and Hamidreza Nazaripouya, propose a "Human-on-the-Loop" (HOTL) control system. Think of it like a video game where an AI plays the game, but a human player watches the screen and can press a special button to change the rules if things get weird. In their system, a computer controller handles the split-second math to keep the voltage steady, but a human supervisor watches the bigger picture. If the inverter starts to degrade (get weaker), the human can step in to tell the computer, "Hey, we're losing power, let's be a little less perfect right now so we don't run out of battery completely." The paper uses simulations to show that this team-up between human judgment and computer speed keeps the grid stable even when the equipment is failing, whereas older, fully-automated methods often crash or run out of power.

The Problem: When the Grid's "Muscles" Get Weak

Imagine your arm is the inverter, and the electricity it pushes is the weight you are lifting. Normally, your arm is strong and can lift exactly what you need. But what if your arm gets injured and can only lift half the weight? If you try to lift the same heavy weight anyway, your arm might lock up, or you might drop the weight entirely. In the world of power grids, "actuator degradation" is like that injured arm. It happens when inverters lose their ability to generate power due to physical damage, cyber-attacks, or just wear and tear.

Old-school solutions try to handle this in two ways, and both have flaws. The first is "Fault-Tolerant Control" (FTC). This is like a robot that has a checklist of every possible injury. If it sees a symptom, it looks up the injury and switches to a pre-programmed fix. The problem? If the injury is weird, new, or happens slowly (like a gradual weakening), the robot might not recognize it, or it might try to fix it too aggressively, causing a crash. The second approach is "Adaptive Control," where the system learns on the fly. But if the system thinks it can still lift the full weight when it actually can't, it might try to push too hard, leading to a loss of control or "parameter drift," where the system gets confused and unstable.

The New Idea: A Human Coach with a "Resilience Dial"

The authors propose a system where a human operator acts as a coach, sitting on the sidelines of the automated controller. This isn't about the human doing the math; the computer still does the heavy lifting. Instead, the human provides "situational awareness." They look at the data and say, "Okay, the inverter is 50% weaker than usual. Let's adjust our strategy."

To make this work, the paper introduces two special tools:

1. Generation Reserve Capacity (GRC): Think of this as your "emergency savings account" of power. It's the extra power the inverter could use if a new emergency happens. The goal is to keep this account full.
2. Controlled Performance Degradation (CPD): This is the "allowable slip-up." Sometimes, to save the emergency savings account, you have to accept that the voltage won't be perfectly steady. It's like deciding to drive a little slower to save gas for a longer trip.

The human operator uses a special knob called $\mu$ (mu) to balance these two. If the operator turns $\mu$ up, the system becomes very conservative. It keeps a huge safety margin (high GRC) but accepts that the voltage might wiggle a bit more (higher CPD). If they turn $\mu$ down, the system tries to be perfect (low CPD) but risks running out of reserve power.

How It Works in Practice

The paper tested this idea using a computer simulation of a grid-connected inverter. They created a scenario where the inverter got "sick" in two stages:

1. At 100 seconds: The inverter lost 50% of its power capacity.
2. At 150 seconds: It lost another 25%, making it very weak.
3. At 200 seconds: A sudden spike in demand (a load disturbance) hit the system.

In the simulation, the human operator watched the screen. When the first 50% loss happened, the operator adjusted the system to handle the drop. But when the second hit came at 150 seconds, the operator realized the situation was getting dangerous. They turned the $\mu$ knob up to 10. This told the computer: "Stop trying to be perfect. Be safe. Keep a big reserve of power so we don't crash."

The result? The system accepted a slightly less perfect voltage reading (the CPD) but successfully kept the lights on and the voltage within safe limits. The inverter never ran out of power, and it didn't get stuck in a "saturation" state (where it tries to push more than it can and fails).

What Happened to the Old Methods?

The researchers compared their new Human-on-the-Loop system against a standard "Fault-Tolerant Control" (FTC) system.

• The FTC System: When the first 50% loss happened, the FTC did okay. But when the second hit came at 150 seconds, the FTC got stuck. It tried to maintain perfect voltage, but because it didn't have enough "muscle" left, it failed. By the time the load spike hit at 200 seconds, the voltage dropped below 0.9 (a dangerous level), and the system couldn't recover.
• The HOTL System: Because the human operator had turned up the safety dial ($\mu$), the system gracefully accepted a small drop in performance. It preserved its reserve capacity, allowing it to handle the final load spike without crashing. The voltage stayed within the safe operating range.

The Bottom Line

This paper doesn't claim to have solved every problem in the world, nor does it say this system is perfect in real life yet. The results are based entirely on simulations—computer models of how the system would behave. However, these simulations suggest that adding a human supervisor to the mix creates a much more resilient grid.

The key takeaway is that sometimes, the best way to handle a crisis isn't to have a robot that never makes a mistake, but to have a robot that knows when to ask a human for advice. By letting a human decide how much "perfection" to sacrifice in exchange for "safety," the system can survive severe damage that would knock out older, fully automated methods. The authors prove mathematically that this approach is stable and that the human's judgment directly helps the system avoid running out of power. It's a reminder that in a complex, high-stakes world, the combination of human wisdom and machine speed might be the strongest defense of all.