Distribution Transformer Parameters Detection Based on Low-Frequency Noise, Machine Learning Methods, and...

Area of Science:

• Electrical Engineering
• Acoustics
• Machine Learning

Background:

• Accurate identification of distribution transformer (DT) parameters is crucial for grid management and maintenance.
• Remote sensing of DT operational characteristics is desirable for safety and efficiency.
• Existing methods may lack accuracy or require direct access to transformers.

Purpose of the Study:

• To develop an automated method for detecting distribution transformer (DT) model, type, and power remotely.
• To utilize low-frequency noise spectra and advanced computational techniques for parameter identification.
• To enhance the accuracy and efficiency of transformer diagnostics.

Main Methods:

• Frequency spectra of sound pressure levels from operating DTs were recorded.
• A hybrid approach combining evolutionary algorithms (genetic algorithm, particle swarm optimization) and machine learning was employed.
• Background noise characteristics were incorporated to account for varying operational conditions.
• Interval selection was validated using five state-of-the-art machine learning algorithms.

Main Results:

• The proposed method achieved high detection accuracies: >84% for model, >99% for type, and >87% for power.
• Genetic algorithm optimization improved accuracy by up to 5% and reduced data input by 80-98%.
• Five machine learning algorithms were identified as effective for this classification task.

Conclusions:

• The developed method offers a reliable and accurate approach for remote, non-invasive identification of distribution transformer (DT) parameters.
• The integration of evolutionary algorithms and machine learning significantly enhances diagnostic capabilities.
• This technique holds potential for improving asset management and predictive maintenance in power systems.