Table 2 The proposed method differs and improves upon existing work.
From: Power system security and protection considering the integration of new energy power plants
Aspect | Existing approaches | Proposed scheme |
|---|---|---|
Noise robustness | Varies; many degrade at low SNR | Maintains >96% accuracy at SNR = 20 dB |
Computational complexity | Deep models (CNN, BiLSTM) often heavy | Lightweight dual-algorithm (AMF + SVM) with fast execution |
Feature extraction | Hand-crafted or deep-learned features | Adaptive Median Filter (AMF) for real-time pre-processing |
Training data dependency | Requires >70% dataset typically | Performs well with only 50% training data |
Response time | Some exceed 20–50 ms | Achieves <15 ms response time |
Accuracy on mixed PQDs | Mixed results; prone to overlap | Consistently detects and classifies individual and combined PQDs |
Real-time suitability | Questionable for many DL-based models | Validated under noisy, dynamic grid with low-latency results |
Interpretability | Limited for deep models | Transparent SVM classification logic |
Generalizability | Often tested on synthetic datasets | Validated across multiple operating conditions |
