Table 9 Comparison with non-AI methods.
From: Bayesian-optimized LSTM-DWT approach for reliable fault detection in MMC-based HVDC systems
Protection method | Detection method | Fault resistance | Noise issue | Sampling rate (kHz) | Communication link | Calculation rate | Other remarks |
|---|---|---|---|---|---|---|---|
Ref.43 | Overcurrent criteria | Low (5.0 Ω) | Yes | 100.0 | No | Medium | Limited performance in long-distance HVDC transmission |
Ref.44 | Capacitor-discharging | Low (10.0 Ω) | Yes | 2.0 | Yes | Higher | In a detailed study, we set the fault resistance to zero |
Ref.45 | DWT | Medium (400.0 Ω) | Yes | 500.0 | No | Medium | The sampling rate is very high |
Ref.46 - | Voltage change-rate | High | No | 200.0 | No | Lower | Further scrutiny is required to examine bias and the challenges of managing high sampling rates |
Ref.47 | Threshold barriers via voltage/current signals | Not discussed | Yes | 50.0 | Yes | Higher | Detailed modelling and precise measurements are required to determine a threshold boundary. These models are susceptible to noise signals generated by the measuring instrument |
Ref.48 | The voltage across bus terminals serves as the basis for establishing threshold barriers | Low (10.0 Ω) | No | 200.0 | Yes | Higher | The classifier must be trained again and reset if some modifications occur in the system suitable for the multivendor application |
Ref.49 | Naïve Bayes classifier | Low (10.0 Ω) | NA | 10.0 | Yes | Higher | While the operating speed is satisfactory, further investigation regarding high fault resistance is needed |
Ref.50 | Current differential | Medium (300.0 Ω) | NA | 10.0 | Yes | Small | High fault resistance is studied successfully; unwanted delays to avoid sensitivity and maloperation need consideration |
Ref.51 | Current signal | Medium (200.0 Ω) | No | 50.0 | No | Medium | A large time window slows down the reaction time |
Proposed method | Improved LSTM | High (480.0 Ω) | No | No | Low | Read non-AI based methods section |
