A power quality disturbances classification method based on multi-modal parallel feature extraction.

Area of Science:

• Electrical Engineering
• Artificial Intelligence

Background:

• Power quality disturbances (PQDs) threaten stable power system operation.
• Traditional methods struggle with complex PQDs due to single-feature focus and high parameter counts.

Purpose of the Study:

• To develop a multi-modal model for accurate and efficient PQD classification.
• To address limitations of traditional single-modal approaches in feature extraction.

Main Methods:

• A parallel feature extraction model combining Long Short-Term Memory (LSTM) for temporal features and a lightweight Residual Network (LResNet) for spatial features.
• Fusion of temporal and spatial features into multi-modal spatio-temporal features (MSTF).
• Classification of MSTF using a Support Vector Machine (SVM).

Main Results:

• Achieved a classification accuracy of 99.94% on 20 PQD signals.
• Significantly reduced model parameters to 0.08 MB.
• Improved accuracy by 2.55% and reduced parameters by 99.25% compared to ResNet18.

Conclusions:

• The proposed multi-modal model effectively extracts spatio-temporal features for accurate PQD classification.
• The lightweight design offers a significant reduction in parameters, enhancing efficiency.
• This approach provides a superior alternative for identifying diverse PQD types in modern power systems.