Deployment of an Online Feedback Optimization Controller for Reactive Power Flow Optimization in a Distribution Grid

summary

The gist

Deployment of an Online Feedback Optimization Controller for Reactive Power Flow Optimization in a Distribution Grid is presented as a successful demonstration of an OFO controller deployed in a real…

In short

An Online Feedback Optimization (OFO) controller was successfully deployed in a real distribution grid for 24/7 operation, reaching Technology Readiness Level 7. This controller optimizes reactive power flow at substations to satisfy voltage constraints while minimizing cost. It proves that OFO can be robust and compatible with existing legacy grid infrastructure.

Key concepts

Online Feedback Optimization (OFO)
A method that solves optimization problems using real-time measurements instead of needing a detailed system model. It works like a feedback loop, using the measured voltage to adjust reactive power injections iteratively until the optimal solution is found.
Reactive Power Flow Optimization
The goal of optimizing how reactive power moves through the grid. The controller adjusts these flows at substations to achieve two main things: minimize operational cost and ensure that voltage levels across the entire distribution network stay within safe limits.
Projected Gradient Descent
A technique used within the OFO controller to figure out how small changes in reactive power injections will affect the desired outcome (voltage). It uses measurements to estimate this effect, allowing the controller to take steps toward the optimal setting without needing a full mathematical model of the grid.
Technology Readiness Level 7 (TRL 7)
A maturity level indicating that a technology has been successfully demonstrated in an operational environment. Reaching TRL 7 means the OFO controller is not just theoretical; it has been proven to work reliably in a real-world, live distribution grid setting.

Terminology used across episodes

This episode discusses

The paper

Deployment of an Online Feedback Optimization Controller for Reactive Power Flow Optimization in a Distribution Grid · Read on arXiv

Lukas Ortmann, Christian Rubin, Alessandro Scozzafava, Janick Lehmann, Saverio Bolognani, Florian Dorfler

Automatic Control Laboratory at ETH Zurich · AEW Energie AG

Optimization is an essential part of power grid operation and lately, Online Optimization methods have gained traction. One such method is Online Feedback Optimization (OFO) which uses measurements from the grid as feedback to iteratively change the control inputs until they converge to the solution of the optimization problem. Such algorithms have been applied to many power system problems and experimentally validated in lab setups. This paper implements an OFO controller in a real distribution grid for 24/7 operation using off-the-shelf hardware and software. The proposed control strategy optimizes the reactive power flow at the substation while satisfying voltage constraints. As part of an existing coordination scheme between (sub)transmission grid operator (TSO) and distribution grid operator (DSO), this comes with a financial reward and simultaneously it virtually reinforces the grid by regulating the voltage on the feeder and therefore allowing higher levels of distributed generation/consumption. We present how a distribution grid is retrofitted such that we can use existing inverters, we analyze the controller's interaction with legacy infrastructure, and investigate its overall control behavior. Finally, we demonstrate the successful deployment of an OFO controller in an operational environment which corresponds to Technology Readiness Level (TRL) 7.

DOI: 10.1109/ISGTEUROPE56780.2023.10408057

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: I'm Rosa.

Dev: And I'm Dev, here with Taro.

Rosa: We're looking at the paper: "Deployment of an Online Feedback Optimization Controller for Reactive Power Flow Optimization in a Distribution Grid".

Dev: Here's the gist of the paper. Deployment of an Online Feedback Optimization Controller for Reactive Power Flow Optimization in a Distribution Grid is presented as a successful demonstration of an OFO controller deployed in a real distribution grid for 24/7 operation, reaching Technology Readiness Level 7 The proposed control strategy optimizes the reactive power flow at the substation while satisfying voltage constraints This deployment serves as a robustness test of OFO and is compatible with legacy grid infrastructure The Optimization Problem Context The core problem addressed is solving an optimization problem that defines a control objective as an optimization problem, such as Optimal Power Flow The specific constraint optimization problem involves minimizing cost minus reward subject to hardware limits and voltage constraints The goal of the OFO controller is to iteratively change the reactive power injections until they converge to q⋆ that optimally solves this problem The Online Feedback Optimization Mechanism Online Feedback Optimization (OFO) is a method used to solve optimization problems using measurements instead of models, functioning as a feedback control method The advantage is that a system model to evaluate v = h(q, d) is not needed, eliminating the need for cable and line parameters or topology knowledge The controller relies on the gradient operator ∇qh(q, d), which describes how a small change in reactive power injections q will change the voltage v Controller Implementation Details The OFO controller is implemented by adapting an integral controller based on projected gradient descent to handle constraints The control input is adapted to the power factor cos(ϕ) of the power injections, which requires enforcing integer constraints due to hardware limits The update law used is cos(ϕ)(k + 1) = cos(ϕ)(k) + αs(cos(ϕ(k)), v(k)) Deployment and Results in the Grid The system was retrofitted to utilize existing hardware, including off-the-shelf inverters and a SCADA system for communication The controller received measurements from the SCADA system through an Archive Data Server once every minute Initial results showed that the inverters did not track the setpoint well, especially for low active power injections, due to old norms requiring reactive power tracking only for larger active power generations The controller automatically adjusted setpoints when the cost function changed based on time-of-day pricing schemes Future Directions and Robustness Possible extensions include state estimation to control voltages not directly measured and incorporating time-varying constraints which can be dynamic line ratings or temporary blocking of taps The paper concludes that OFO is robust against model mismatch, compatible with legacy infrastructure, and can work with triggered measurements Further investigations are needed to quantify the monetary value of the virtual grid reinforcement provided by voltage control The implementation demonstrates that OFO has reached TRL 7 REFERENCES

Title and authors: Rosa: The paper we're discussing is "Deployment of an Online Feedback Optimization Controller for Reactive Power Flow Optimization in a Distribution Grid," written by Lukas Ortmann, Christian Rubin, Alessandro Scozzafava, Janick Lehmann, Saverio Bolognani, and Florian Dorfler.

Dev: That's a whole team behind it. The title really tells you what this is about: taking an Online Feedback Optimization controller and deploying it to optimize reactive power flow in a distribution grid. It sounds complex, but the core idea is using measurements to guide control adjustments in real-time to solve an optimization problem.

Taro: When you say deployment, does that mean they just plugged it into some existing setup, or did they have to build something custom from scratch for this grid? I'm thinking about how robust it is when the hardware isn't brand new.

Rosa: They actually retrofitted an existing system, using off-the-shelf inverters and a standard SCADA system that was already in place for communication. The key point is that this deployment serves as a robustness test for the OFO method itself, showing it can handle the messiness of a live grid.

Dev: And they achieved technology readiness level seven, or TRL seven which means they’ve proven it works in an operational environment for twenty-four hours a day <ref:2305.06702#pg1>. That’s a big deal when you think about putting complex control logic into something that has to run reliably constantly.

The paper's summary: Rosa: So, the paper summarizes that the control strategy they proposed optimizes the reactive power flow at the substation while making sure they satisfy all those voltage constraints. It’s built around solving an optimization problem where you’re trying to minimize a cost function minus some reward, subject to hardware limits and those voltage constraints.

Dev: They explain that the goal of this Online Feedback Optimization controller is to iteratively change the reactive power injections until they converge onto that specific value, q star, which is what optimally solves that initial problem. It’s basically an iterative search process driven by feedback loops.

Taro: Iterative search sounds like it could get stuck in a local optimum if the grid conditions change really fast or if there's some unexpected disturbance. What safeguards did they put in place to make sure this iterative process actually finds the right answer, q star?

Rosa: They use a method called projected gradient descent to handle those constraints, which is how they ensure that the reactive power injections and the resulting voltages stay within those required limits during every update step. That’s what keeps it stable.

Dev: And they show an illustrative example of this idea using standard gradient descent, where you take steps with a size alpha based on the gradient of the cost function, but they explicitly point out that standard gradient descent doesn't satisfy those constraints on reactive power injections and voltages at all.

The paper's improvements: Rosa: The authors point out several ways this approach could be enhanced. One improvement involves adaptive setpoint tracking for the inverters, which they say can be better if you build a control law that explicitly accounts for actuator limitations, like when there are low active power injections causing issues.

Dev: That sounds like addressing the problem where the inverters don't track their targets well, especially when there are low active power injections because of some old norms requiring them to only track reactive power for bigger active power generations. So they want better handling of those specific hardware quirks.

Taro: I'm interested in the idea of incorporating state estimation into the controller, because the paper mentions controlling voltages that aren't directly measured. That could help if a certain sensor fails or if you just can't get a direct reading for some buses.

Rosa: Exactly, they suggest integrating state estimation so you can control those voltages even when they aren't directly measured, which would allow for better voltage regulation in more complex scenarios where direct measurements are unavailable. They also look at incorporating time-varying constraints later on.

Dev: Time-varying constraints would be useful for situations where the rules change, like if you have dynamic line ratings or temporary blocking of taps changers, allowing the controller to reduce power flow on a specific line based on what’s happening right now. It shows they are thinking about making it more flexible.

Conclusion: Rosa: So, to wrap up the paper "Deployment of an Online Feedback Optimization Controller for Reactive Power Flow Optimization in a Distribution Grid," the main thing is that this OFO method reached TRL seven by being deployed in a real distribution grid for twenty-four hours a day and seven days a week <ref:2305.06702#pg1,Deployment of an Online Feedback Optimization Controller for Reactive Power Flow Optimization>.

Dev: Essentially, it provides a robust way to optimize reactive power flow at the substation while strictly satisfying voltage constraints, even when you're using existing hardware and just relying on measurements from the SCADA system.

Taro: For someone just listening to the show, what does this mean practically? It means we have a method that doesn't need a perfect model of the grid to keep things running smoothly in the real world, even with some inherent measurement noise or uncertainty.

Rosa: It shows that OFO is robust against model mismatch and it’s compatible with older grid infrastructure, which is really important because you don't always have the luxury of installing brand new equipment everywhere.

Dev: The paper confirms that this method works with triggered measurements too, which means it can adapt when the data stream changes dynamically, pushing the boundaries of what we thought was feasible for these types of control problems.

Taro: It’s promising because it moves beyond just theoretical lab results and shows real-world operational capability, which is where I spend a lot of time testing failure modes.

Rosa: That’s the summary on this paper, "Deployment of an Online Feedback Optimization Controller for Reactive Power Flow Optimization in a Distribution Grid." We’ll be looking at some other papers next time to see how these control concepts are applied elsewhere.

More episodes

← Home