Table 6 Distinctive features of investigated knowledge-based protection schemes
References | Year | Citation | DER | Required measures | Method | Features |
|---|---|---|---|---|---|---|
[131] | 2017 | 216 | Synchronous and inverter based | Current data | ANN with WT | Robust against measurement noise and uncertainty High computational burden due to training process Accuracy varies with system configuration |
[132] | 2019 | 74 | Synchronous and inverter based | Voltage and current signals | ANN-SVM-based overcurrent and distance protection | Adaptive. Self-learning, and self-training High computational burden Complex implementation High accuracy |
[133] | 2019 | 13 | Inverter based | Current data | ANN | Uses TMF to discern temporary/permanent faults. Improves auto-recloser functionality by discriminating permanent/transient failures. Requires less computing time Online training is feasible |
[134] | 2018 | 14 | Inverter based | Current data | Fuzzy logic | Two Fuzzy logic models for firstly deciding operating mode, and then detecting/classifying internal faults of MG Response time is about 0.25 – 1 cycle Simple and feasible implementation Robust against DERs outages and load variation |
[135] | 2015 | 33 | Synchronous and inverter based | Current data | Fuzzy logic and decision tree | High computational burden due to decision tree training Large number of extracted features Response time is about 2.25 cycle High impedance faults are detectable |
[136] | 2018 | 46 | Inverter based | Voltage and current signals | Type-2 Fuzzy logic | Addresses data uncertainties Identifies fault and its direction Low computational burden No need for training |
[139] | 2018 | 18 | Inverter based | Voltage and current signals | Bagged decision tree | Considers changes in load, generation, and fault resistance Applicable for high impedance faults Robust against data noise Large dataset for tree training High computational burden due to training |
[141] | 2017 | 20 | Synchronous and inverter based | Voltage and current signals | SVM and WT | Considers changes in fault resistance, location, and initiation timing High computational burden due to training process Applicable for high impedance faults Fault classification accuracy nears 95.5% |