SaeroSense새로센스
Start a project
  1. Home
  2. Work
  3. Auto-reclosing with particle filter & CNN

Case study · Network & telemetry AI

Auto-reclosing with particle filter & CNN

A single-phase auto-reclosing scheme for shunt- and non-shunt-compensated transmission lines. A particle filter estimates voltage harmonics, two engineered indices capture arc behavior and fault type, and a CNN decides whether — and when — breakers can safely reclose.

Published in IEEE Transactions on Power Delivery

Sector
Power systems · transmission
Context
Peer-reviewed research
Practice
Network & telemetry AI
Delivered by
Saerosense founding team
  • 2,442fault cases90% training · 10% testing
  • 2engineered indicesSecondary-arc detection γ₁ and fault recognition γ₂
  • 3detectors comparedCNN, DNN and SVM classifiers

The challenge

After a single-phase fault, breakers must reclose quickly to keep the line in service — but reclosing into a permanent fault, or before the secondary arc has extinguished, damages equipment and destabilizes the grid. Fixed dead-times are either too slow or unsafe, especially on compensated lines.

Our approach

The faulty-phase voltage is decomposed into harmonics with a particle filter. Two indices are computed: γ₁ tracks secondary-arc extinction from harmonic energy, and γ₂ separates temporary from permanent faults. A CNN classifies these features, and breakers reclose only when the arc is extinguished and the fault is temporary.

Workflow

How it works, step by step.

The pipeline in 5 stages, from measure to decide. Each stage has a clear input, a clear output and a reason to exist.

  1. 01

    Measure

    Faulty-phase voltage captured at the line terminal.

    • Voltage waveform
  2. 02

    Estimate

    A particle filter estimates the harmonic components.

    • Particle filter
  3. 03

    Extract

    Secondary-arc index γ₁ and temporary/permanent fault index γ₂.

    • Feature engineering
  4. 04

    Classify

    1-D CNN with 2,348 input nodes trained on 2,442 cases; compared with DNN and SVM.

    • CNN
    • DNN
    • SVM
  5. 05

    Decide

    Arc extinguished and fault removed → reclose. Otherwise breakers stay open.

    • Reclosing logic

Results

What it produced.

The scheme end to end — particle-filter preprocessing, feature extraction and CNN-based decision stage.
The scheme end to end — particle-filter preprocessing, feature extraction and CNN-based decision stage.
Voltage spectrograms and γ₁ amplitude for no-arc, short-arc and prolonged-arc faults — the moment of secondary-arc extinction is clearly visible.
Voltage spectrograms and γ₁ amplitude for no-arc, short-arc and prolonged-arc faults — the moment of secondary-arc extinction is clearly visible.
γ₂ amplitude separates temporary and permanent faults across arc conditions.
γ₂ amplitude separates temporary and permanent faults across arc conditions.
CNN architecture: three convolution layers, pooling, and fully connected layers with dropout.
CNN architecture: three convolution layers, pooling, and fully connected layers with dropout.

Stack & techniques

  • Particle filter
  • Harmonic estimation
  • CNN
  • DNN
  • SVM
  • Power-system signal processing

Delivered by members of the Saerosense founding team. Client names are withheld unless public — references are available on request.

Related service

ServiceNetwork & telemetry AIPortfolioAll case studies

Start a project

Have a similar problem?

Send a few lines about your data and the outcome you want. We reply within two business days with a feasibility read and a proposed scoping plan.

Start a conversation contact@saerosense.com Seoul, South Korea · remote worldwide · English & 한국어