⚡ electrical engineering

Feedback Linearization and Control of a Grid-Forming Power Converter in an Islanded Microgrid

This paper proposes a full-state feedback linearization control strategy for grid-forming inverters in islanded microgrids that achieves significantly faster transient response than traditional cascaded PI controllers by exactly canceling nonlinearities to create independent double-integrator dynamics, albeit with a trade-off in robustness to parameter mismatches.

Original authors: Rene Ebunle Akupan, May-Win Thein, Se Young Yoon

Published 2026-06-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Rene Ebunle Akupan, May-Win Thein, Se Young Yoon

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine an islanded microgrid as a small, isolated island where a group of solar panels and wind turbines must power a village. In this scenario, there is no massive mainland power grid to lean on for stability. The "grid-forming inverter" is the smart manager of this island. Its job is to create a perfect, steady electrical voltage (like a steady water pressure in a pipe) that the village's appliances rely on.

The problem is tricky: The amount of electricity the village uses (the load) changes instantly based on how hard the manager pushes the voltage. If the manager pushes too hard, the village uses more power; if they push too little, the voltage drops. It's a constant, fast-paced tug-of-war.

The Old Way: The "Two-Step" Manager (Cascaded PI)

For years, engineers have managed this using a Cascaded PI Controller. Think of this as a manager with two assistants working in a strict hierarchy:

The Fast Assistant (Inner Loop): This person watches the current flowing through the wires. They react very quickly to keep the current steady.
The Slow Manager (Outer Loop): This person watches the voltage. They tell the Fast Assistant what to do, but they move slowly. They wait for the Fast Assistant to finish their job before making a new decision.

To make this work, the Slow Manager has to be much slower than the Fast Assistant. If they aren't, the system gets confused and unstable. Also, because the electricity rotates (like a spinning wheel), the two directions of current (d-axis and q-axis) tend to mess with each other. The old system tries to fix this mess with "feedforward" guesses—like a driver guessing how much to turn the steering wheel based on a map. But this guess only works perfectly if the road conditions (the load and resistance) are exactly what the map says. If the road changes, the guess is wrong, and the car drifts.

The Result: This system is stable but sluggish. When you ask it to change the voltage, it takes a long time to settle down.

The New Way: The "All-Knowing" Manager (Feedback Linearization)

The paper proposes a new approach called Full-State Feedback Linearization. Instead of using a slow-and-fast hierarchy, this method treats the entire system as a single, unified puzzle that can be solved mathematically.

Here is the analogy:
Imagine the electrical system is a complex, bumpy roller coaster. The old method tries to drive the coaster by gently nudging the wheels and hoping it stays on track. The new method realizes that if you know the exact shape of the track and the physics of the car, you can calculate a perfect steering command that cancels out all the bumps and curves instantly.

The new controller does three magical things:

It cancels the "bumps": It mathematically eliminates the confusing rotation effects and resistance drops that usually make the system hard to control.
It straightens the track: By canceling those bumps, the complex, non-linear system becomes as simple as a double integrator. In plain English, this means the system behaves like a car on a perfectly straight, frictionless highway. If you press the gas, it accelerates smoothly; if you let go, it stops smoothly. There are no hidden surprises or internal wobbles.
It acts instantly: Because the system is now "straight," the controller can use a standard, high-speed strategy (pole placement) to get the voltage to the target almost immediately.

The Showdown: What Happened in the Lab?

The authors tested both managers in a computer simulation (MATLAB) under three specific challenges:

1. Changing the Target (Reference Tracking)

The Task: Suddenly change the voltage target from 359V to 320V.
The Old Manager (PI): It took more than 50 milliseconds and hadn't even finished the job by the time the test ended. It was still drifting toward the target.
The New Manager (FL): It hit the target in just 0.76 milliseconds. It was over 65 times faster.

2. A Sudden Load Spike (Load Step)

The Task: Suddenly double the power demand (like turning on a massive factory).
The Old Manager: The voltage dipped and took a long time to recover. It was still climbing back up after 50ms.
The New Manager: The voltage dipped slightly (because the physics of the inductor couldn't change instantly), but it recovered to the target in about 5 milliseconds. It was 10 times faster at recovering.

3. The "Wrong Map" Test (Parameter Mismatch)

The Task: The new manager was given a slightly wrong value for the wire resistance (a 50% error).
The Old Manager: It didn't care. Because it uses a "slow and steady" integral action, it eventually corrected itself regardless of the wrong map.
The New Manager: Because it relies on a perfect mathematical cancellation, the wrong map caused a tiny, permanent error (about 1.3 volts off). It was very fast, but slightly less forgiving of bad data.

The Bottom Line

The paper concludes that the new Feedback Linearization controller is a structural upgrade, not just a tuning tweak.

The Trade-off: You get blazing speed and perfect separation of electrical directions (no cross-talk) with the new method. However, you lose a tiny bit of "forgiveness" if your hardware measurements are slightly off.
The Verdict: If you know your hardware well (which is usually true for engineered systems), the new method is vastly superior for speed and stability. If your hardware parameters drift wildly over time, the old, slower method might still be safer.

In short, the new controller turns a chaotic, bumpy ride into a smooth, high-speed rocket, provided the pilot has an accurate map.

Technical Summary: Feedback Linearization and Control of a Grid-Forming Power Converter in an Islanded Microgrid

Problem Statement
In islanded microgrids, grid-forming inverters face a fundamental control challenge: they must regulate terminal voltage and frequency without an external grid reference, despite the fact that the load current they supply is directly dependent on the voltage they generate. The standard industry approach utilizes a cascaded proportional-integral (PI) controller with a fast inner current loop and a slower outer voltage loop. While widely used, this architecture relies on bandwidth separation and feedforward compensation for $dq$ rotational coupling. These mechanisms are only exact at the specific operating point where the controller is tuned, limiting the system's response speed and robustness to changing conditions. Furthermore, the cascaded structure does not fully exploit the mathematical properties of the inverter model, which is input-affine and admits exact linearization.

Methodology
This paper proposes a full-state feedback linearization (FL) controller as an alternative to the cascaded PI approach. The methodology proceeds as follows:

System Modeling: A four-state nonlinear model of the islanded grid-forming inverter is developed in the synchronous $dq$ reference frame. The model includes the LC output filter dynamics, cross-coupling terms induced by the rotating frame, and the algebraic constraint of a purely resistive load.
Relative Degree Analysis: The authors demonstrate that the system possesses a vector relative degree of $(2, 2)$ , where the sum of the relative degrees equals the system order (4). This confirms that the system admits full-state linearization with no internal or zero dynamics.
Controller Design:
- Linearization: A single feedback law is derived to cancel the system's nonlinearities exactly. This includes rotational coupling ( $\omega^* L_f i_q$ , etc.), resistive drops ( $R_f i_d$ ), and load conductance terms.
- Decoupling: The resulting closed-loop input-to-output map transforms the nonlinear system into two independent double integrators.
- Pole Placement: A standard linear pole-placement design is applied to the virtual control input to shape the transient response, allowing for arbitrary placement of closed-loop poles.
Implementation and Comparison: The proposed FL controller is implemented in MATLAB and compared against a cascaded PI baseline (tuned according to established bandwidth separation procedures) under identical conditions at a 20 MW operating point. Three scenarios are tested: reference tracking, load step disturbances, and parameter mismatch.

Key Contributions and Results
The paper presents a comparative analysis highlighting the structural differences between the feedback-linearizing controller and the cascaded PI baseline:

Transient Performance: The FL controller demonstrates significantly faster transient response. In reference tracking (a 39 V step), the FL controller settles within 0.76 ms, whereas the PI controller fails to reach the 2% settling band within the 50 ms simulation window. Similarly, for a load step doubling the power from 20 MW to 40 MW, the FL controller recovers in approximately 5 ms, compared to the PI controller's slow, incomplete recovery.
Decoupling: The FL controller achieves machine-precision decoupling between the $d$ and $q$ axes, with peak $q$ -axis cross-coupling below 1 mV. In contrast, the PI controller exhibits 41 mV of cross-coupling due to imperfect feedforward compensation.
Robustness Trade-off: The paper acknowledges a trade-off regarding parameter uncertainty. The cascaded PI controller exhibits superior robustness to filter parameter mismatch (specifically a 50% error in filter resistance $R_f$ ), maintaining zero steady-state error due to its inner-loop integral action. The FL controller, which relies on exact model cancellation, exhibits a small steady-state offset (1.3 V) when the model parameters do not match the plant. However, the authors note that this offset is small because the filter resistance itself is small.

Significance and Claims
The paper positions this work as a tutorial demonstrating that the performance limitations of the standard cascaded PI approach are structural rather than tuning-related. The primary significance lies in showing that:

Exact Linearization is Feasible: The islanded inverter model allows for exact state-space linearization, eliminating the need for bandwidth separation and operating-point-dependent compensation.
Structural Advantages: The superior speed and decoupling of the FL controller are direct consequences of its architecture, which cancels nonlinearities rather than suppressing them with high gains.
Design Philosophy: The choice between FL and PI depends on the specific constraints of the application. If transient performance is the binding constraint and filter parameters are known within manufacturing tolerances, the FL architecture is superior. If filter parameters drift significantly over the system's lifetime, the integral action of the cascaded PI offers a robustness advantage.

The authors conclude that while the FL controller offers dramatic performance improvements, future work could address the robustness trade-off by adding integral action to the outer linear loop of the FL controller, thereby recovering robustness to constant parameter errors without sacrificing the benefits of feedback linearization.

The Old Way: The "Two-Step" Manager (Cascaded PI)

The New Way: The "All-Knowing" Manager (Feedback Linearization)

The Showdown: What Happened in the Lab?

The Bottom Line

Technical Summary: Feedback Linearization and Control of a Grid-Forming Power Converter in an Islanded Microgrid

More like this