Supervised relevance vector machine based dynamic disturbance classifier for series compensated transmission line.

Absrtact: Background: Nowadays, numerical relays configured through power swing blocking function are vulnerable to faults with high impedance owing to nonlinearities present in the faulty path. Identification of symmetrical high impedance faults incepting during the power swing scenario is of prime importance for protection engineers to avoid cascade tripping. It is analyzed that the classical approaches for symmetrical fault detection during power swing can mal‐operate owing to high level of uncertainty arise during validation of high impedance fault, symmetrical faults and cross country faults during power swing with varied level of compensation which has been explored in this research. Aims: This paper presents an innovative scheme for series compensated line by utilizing the capability of Relevance Vector Machine (RVM) as a disturbance classifier taking into consideration of cross country faults, CT saturation; power swings prone to symmetrical faults along with high impedance faults in comparison with prevailing methods. Materials & Methods: In order to validate the proposed RVM based disturbance classifier algorithm, different kinds of faults and power swing situations have been applied in IEEE 9 bus test system using PSCAD/EMTDC software followed by validation in MATLAB. The features of sequence components of currents are estimated using Modified Full Cycle Discrete Fourier Transform (MFCDFT). The extracted features are used as control inputs to RVM for discernment among power swing and fault under the influence of wide range of system and fault dynamics. Results: The developed RVM based disturbance classifier scheme generates a tripping signal in the event of high impedance fault during power swing and remains stable during swing conditions. It can be used in conjunction with an adaptive numerical distance relaying scheme for realizing complete protection of the power grid for all zones of protection. A comparative analysis with an existing disturbance classifier algorithm indicates relative superiority with much improved fault classification accuracy. Discussion: In order to validate the proposed algorithm, different kinds of fault and power swing scenario are realized in IEEE 9 bus test system using PSCAD and current samples are acquired at sampling frequency of 4 kHz. The phasor estimation of sampled signals has been performed using MFCDFT which is used to derive feature vector extraction from symmetrical components. The extracted features are applied to RVM based fault classifier algorithm for classification of disturbance. The obtained results are also compared with existing schemes of protections. It has been proved that developed algorithm outperforms with much improved fault classification accuracy. Conclusion: This paper discovers relevance vector machine based adaptive learning approach for dynamic disturbance classification in complex power grid network. The developed algorithm has been validated on IEEE 9 bus test system modelled in PSCAD. The fault and system dynamics are varied over a wider range which includes variation in fault type, fault impedance, fault inception angle, power flow angle, fault location and degree of line compensation level. Moreover, the validation of scripted algorithm has been performed for very complex power grid scenarios like cross‐country faults, high impedance faults with varying degree of compensation, power swing and saturation of instrument transformers etc. The feature vectors are extracted using MFCDFT and applied to supervised RVM based disturbance classifier and outperforms with classification accuracy of 99.74% which indicates it's superiority over existing classifier approaches based on SVM & ANN. [ABSTRACT FROM AUTHOR]