Modified Communication Assisted Line Differential Protection Scheme with Adaptive Machine Learning-Based Relay

Although line differential protection schemes have superior fault sensitivity and speed of operation, their dependency on pilot signals introduces certain challenges. These challenges include nuisance trips due to intolerably high communication delay and loss of fault sensitivity in the event of communication link failure. Also, to incorporate potential mal-operation in the event of CT saturation, line charging current and marginal delay, higher restraining zones are set for line differential relay. Hence, sensing evolving faults with higher impedances is difficult with the differential principle. Considering these limitations, a novel Machine Learning (ML) based approach is proposed in the article to assist differential relay in detecting high impedance faults and allow larger restraining zones. The proposed ML-based relay also acts as a backup in case of communication link failure. The ML models used such as Decision Tree, Support Vector Machines, and K-Nearest Neighbors also adapt their training to detect each fault. To improve computation speed and reduce calculation burden, a novel ensemble of dimensionality reduction using PCA, Linear Regression, and Pearson Coefficient is presented in the article. Data for ML models are obtained and validated using MATLAB and PSCAD. Data pre-processing and algorithm testing are done in Python.