Hybrid Deep LSTM-XGBoost Based Fault Line Identification and Distance Estimation in Grids with Distributed Generators

Worldwide, electricity use is growing in popularity to promote sustainable human living and economic prosperity. Human existence is extremely impossible without power, and the nation's ability to function normally will collapse. In order to meet the world's power demand, distributed renewable energy generator integration is growing rapidly these days. Power systems frequently experience faults, which are not totally prevented. However, sophisticated techniques mitigate the fault quickly, allowing the power system to recover. Two difficult yet essential tasks for the power system to function continuously are fault distance estimation and fault line identification. Because distributed energy sources integrate dynamically and have bidirectional current flow with a low magnitude of fault current, fault diagnosis becomes more difficult. Although machine learning models have gained popularity recently, feature engineering on them requires manual preprocessing. This study proposes a hybrid of deep learning and ensemble learning models, i.e., Deep Long-sort term memory and Extreme Gradient Boosting (LSTM-XGBoost), for fault line identification and fault distance estimation tasks. Two different distributed generators have been used to perform this task: a Doubly Fed Induction Generator (DFIG) wind turbine and a diesel generator. Different fault configurations have been created, including resistances, initiation angles, and locations. A hybrid of the Deep LSTM-XGBoost model gives better 99.58 % accuracy and 0.0045 MAE for the intended tasks. Various comparisons with existing deep learning models from literature surveys have been discussed. Hardware in Loop (HIL) simulation has been performed to validate the robustness and flexibility of the proposed model using the OPAL-RT 4510 real-time simulator.