Directionally Supervised Neuromorphic Protection for Distribution Systems Using Local Measurements

summary

The gist

Reliable protection under varying power system operating conditions requires rapid fault detection while preserving selectivity against external faults and non-fault disturbances This paper proposes…

In short

The study developed a communication-free protection scheme for power systems that detects internal faults while ignoring external disturbances. It uses local measurements to create an index for disturbance severity and directional information to distinguish between fault directions, employing a neuromorphic system with adaptive firing thresholds. This method proved feasible in real-time tests across various network types.

Key concepts

Disturbance Index (D[k])
This is a single number calculated by adding up the normalized deviations of voltage, current, active power, and reactive power from their normal values. If this index exceeds a set limit, it signals that an event is happening in the system.
Directional Index (Γ[k])
This mechanism analyzes changes in positive-sequence voltage and current to determine the direction of a fault. It uses admittance indicators to calculate a signed value that tells the system whether the disturbance is moving in a forward or reverse direction.
Adaptive Firing Threshold (Θ[k])
This threshold for an artificial neuron changes based on how severe the disturbance is and its direction. This allows the system to be sensitive enough to detect small disturbances while remaining selective against external noise and correctly identifying internal faults.

Terminology used across episodes

This episode discusses

The paper

Directionally Supervised Neuromorphic Protection for Distribution Systems Using Local Measurements · Read on arXiv

Saurabh Prabhakar, Yubo Song, Pascal Schirmer, Bijaya Ketan Panigrahi, Subham Sahoo

Reliable protection under varying power system operating conditions requires rapid fault detection while preserving selectivity against external faults and non-fault disturbances. This requirement is particularly challenging in communication-free schemes, where relays must determine trip conditions solely from local electrical measurements. To address this challenge, this paper proposes a directionally supervised neuromorphic protection scheme that integrates disturbance-responsive leaky integrate-and-fire dynamics with a direction-dependent firing threshold to distinguish internal faults from reverse-direction events. The unified relay architecture combines local disturbance detection, directional discrimination, and spike-based trip generation without requiring inter-relay communication. Its performance is evaluated under grid-connected and islanded conditions using a ring microgrid and the IEEE 13-node distribution feeder. Across 240 evaluated fault scenarios, the framework achieves median operating times of 5.4--7.6 ms across the two networks, with 78% of cases operating within 10 ms and 95% within 20 ms. It also maintains protection security during load disturbances, while fault-resistance analysis confirms its response under high-resistance faults. Controller-hardware-in-the-loop (CHIL) experiments using an RTDS simulator and an external dSPACE controller demonstrate trip-command assertion in approximately 10 ms and 5 ms for the tested single-line-to-ground and three-phase-to-ground faults, respectively. The results establish real-time feasibility within the evaluated conditions and characterize operating-time variability across fault types and resistance levels for different network configurations.

Transcript

Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.

Rosa: I'm Rosa.

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

Rosa: We're looking at the paper: "Directionally Supervised Neuromorphic Protection for Distribution Systems Using Local Measurements".

Dev: Here's the gist of the paper. Reliable protection under varying power system operating conditions requires rapid fault detection while preserving selectivity against external faults and non-fault disturbances This paper proposes a directionally supervised neuromorphic protection scheme that integrates disturbance-responsive leaky integrate-and-fire dynamics with a direction-dependent firing threshold to distinguish internal faults from reverse-direction events Across 240 evaluated fault scenarios, the framework achieves median operating times of 5.4–7.6 ms across the two networks The results establish realtime feasibility within the evaluated conditions and characterize operating-time variability across fault types and resistance levels for different network configurations.

Rosa: Framework Architecture The unified relay architecture combines local disturbance detection, directional discrimination, and spike-based trip generation without requiring inter-relay communication The proposed framework comprises three interconnected stages.

Dev: First, local voltage, current, active-power, and reactive-power measurements are normalized using their rated bases to construct a multi-domain disturbance index, D

k: , which governs event activation and latencybased spike encoding Second, positive-sequence voltage and current increments are processed through sequence-admittance indicators to determine the signed directional index, Γ

k: , and the corresponding directional state, M

k: .

Rosa: Third, the encoded disturbance spikes are integrated by an LIF neuron, whose adaptive firing threshold, Θ

k: , accounts for disturbance severity and direction.

Paper summary: Rosa: So, we're looking at this paper called "Directionally Supervised Neuromorphic Protection for Distribution Systems Using Local Measurements." Basically, it tackles the problem of needing fast fault detection in power systems without needing any communication between the relays.

Dev: Right. The core idea is to use a neuromorphic approach that combines how much disturbance there is with whether that disturbance is coming from a forward or reverse direction, all using just local measurements.

Rosa: It claims they propose this scheme integrates disturbance-responsive leaky integrate-and-fire dynamics with a direction-dependent firing threshold to separate internal faults from events happening in the wrong direction.

Dev: That sounds like they're trying to build a system that's smart enough to tell the difference between a fault and just some normal power flow change, which is tough when you have bidirectional power flow.

Taro: I wonder how this works in practice when things get messy on the grid; does it handle those unexpected events well, or does it get confused?

Rosa: The paper evaluates this framework across two hundred forty different fault scenarios, and they found median operating times of five point four to seven point six milliseconds across the two networks they tested <ref:2611.16711#pg3>.

Dev: Five to seven milliseconds is pretty fast for protection schemes. It shows they're looking at real-time feasibility under those evaluated conditions.

Taro: But how does that timing change when the fault resistance increases, say from zero point five ohms up to fifty ohms? That’s where I get worried about the performance dropping off.

Rosa: They found operating time increases with fault resistance across all six categories, rising from approximately five to six milliseconds at a fault resistance of zero point five ohms up to twelve to fourteen milliseconds when the fault resistance is fifty ohms.

Dev: So, it means that while it's fast for some faults, the response can get quite slow if the physical condition of the fault changes significantly. That variability across those different fault types and resistance levels is something you have to keep in mind.

Taro: From a real-world testing standpoint, what does this mean for a system that actually has to recover gracefully when things go wrong? Does this directional discrimination help it recover smoothly?

Rosa: The whole framework relies on three interconnected stages. First, they normalize local measurements like voltage and current into a multi-domain disturbance index, which drives the initial event activation and sets the latency based on that index.

Paper summary: Dev: Then they get to directional discrimination, where positive-sequence voltage and current increments are processed through sequence-admittance indicators to determine a signed directional index and the corresponding directional state.

Taro: So it’s not just detecting a drop in voltage, but figuring out if that drop is due to something flowing the wrong way across the network.

Rosa: Exactly. And finally, those encoded disturbance spikes are integrated by a leaky integrate-and-fire neuron that has an adaptive firing threshold that accounts for both the severity of the disturbance and its direction.

Dev: That adaptive threshold modulation is key because it’s what lets it distinguish between a true fault and just noise or a reverse-direction event, which is why they focused on making that dynamic.

Taro: So if the system thinks it's going in reverse, does that mean the firing threshold goes up, making it less sensitive to other disturbances?

Rosa: Yes. When the directional state is set to reverse direction, the neuron's adaptive threshold increases, which acts as a restraint against tripping unless there’s a very severe disturbance present.

Dev: That makes sense for safety; you don't want false trips during normal power flow conditions because of some transient asymmetry in the measurements.

Taro: What about when it fails? The paper mentions that the unified relay architecture combines local disturbance detection, directional discrimination, and spike-based trip generation without needing inter-relay communication. What’s the failure mode there?

Rosa: The paper doesn't explicitly detail a failure mode for a single component, but its strength is that it removes the need for inter-relay communication entirely, which simplifies deployment but also means any local measurement error is directly translated into a trip decision.

Dev: That’s the trade-off—simpler hardware footprint and no network dependencies, but total reliance on the accuracy of those initial local measurements to make a safe trip command.

Taro: Does this mean if there's a faulty sensor reading, say one of those voltage measurements being off, the entire directional discrimination fails?

Rosa: It suggests that the success hinges on how well they calibrate that sensitivity coefficient kappa, which is set by dividing the common threshold Dth by the maximum disturbance Dmax and then squaring the time constant tau max.

Paper summary: Dev: And they calibrated it so that if you hit that maximum disturbance, you get a minimum encoding latency of two times tau max when the condition is right, provided Dmax is greater than Dth and tau max is greater than two times tau sim minus one.

Taro: So it’s a tight calibration requirement for the system to perform at its best speed. What about the hardware in-the-loop testing they did? Did it hold up when they added load changes?

Rosa: In their Controller-hardware-in-the-loop experiments, using an RTDS simulator and an external dSPACE controller, they demonstrated trip command assertion within approximately ten milliseconds for a single line to ground fault and five milliseconds for a three phase to ground fault.

Dev: Those numbers are solid for hardware assertion. And they also showed that across all two hundred forty cases, the median operating times stayed below the ten millisecond reference time.

Taro: But they also noted that cumulative operating-time distribution showed that about ninety five percent of the evaluated cases were resolved within twenty milliseconds, which is a reasonable bound for some contingencies.

Rosa: And one specific finding from their CHIL experiments was that hardware based trip signal assertion happened within approximately five milliseconds specifically for an ABCG fault, and they noted no trip during the tested load addition.

Dev: That load addition test is important because it checks the system’s behavior when things aren't just static faults but dynamic changes in the power flow.

Taro: It seems like this framework is really designed to provide a fast, selective response based on what happens right there at the measurement point, which is exactly what we need for high-speed protection.

Rosa: Ultimately, it proves that you can get fast and selective protection across different microgrid configurations using just local measurements by combining disturbance encoding latency with directional classification.

Dev: It’s a solid demonstration of how neuromorphic dynamics can handle the complexity of bidirectional power flow in a communication-free environment.

Taro: So for someone managing these systems, this implies you can have faster protection without needing complex, slow communication networks to coordinate the relays.

Rosa: That’s the big implication—faster response times achievable locally across diverse grid setups.

Dev: The paper's title, "Directionally Supervised Neuromorphic Protection for Distribution Systems Using Local Measurements," really captures that focus on directionality and locality.

Taro: It seems like this work is moving toward protection schemes that are inherently more resilient to the uncertainties of a dynamic power system by being locally aware.

Conclusion: Rosa: So this paper is about using local measurements to make protection smarter, specifically by figuring out if a fault is going in the right direction or the wrong direction.

Dev: Right, they're proposing this neuromorphic approach that uses a leaky integrate-and-fire dynamic combined with directional detection to avoid needing any communication between relays.

Rosa: They found that across two different network setups, their system could trip in about five point four to seven point six milliseconds on average for the faults they tested.

Dev: That’s fast enough for real-time operation, but they also showed that the response time gets slower as the fault resistance goes up from zero point five ohms all the way to fifty ohms.

Taro: So even though it's fast at low resistance, if you have a high resistance fault, you might wait twelve to fourteen milliseconds for a trip command.

Rosa: Exactly. They also showed that across two hundred forty different scenarios, about ninety five percent of the faults were handled within twenty milliseconds.

Dev: That’s good data for validation, but the real point is how this works when the grid gets weird or when you're just testing it outside of a clean lab setting.

Taro: I think it shows that you can build protection that doesn't rely on talking to other parts of the system to work fast and selectively.

Rosa: It suggests we can get faster, more localized protection for microgrids without needing complex communication networks running everywhere.

Dev: That moves the safety case closer to where the actual hardware is deployed, which is a big step for reliability in these kinds of systems.

More episodes

← Home