Transient stability assessment and enhancement of hydropower plant using artificial neural network

Abstract

Power system stability is essential for a reliable electricity supply, as disturbances such as faults, load changes, and line outages can cause blackouts and generator instability. As Ethiopia’s base-load hydropower plant, the Tana Beles Hydropower Plant (TBHPP) requires rigorous transient stability analysis and enhancement to maintain system reliability after major disturbances. This study tackles these challenges by employing an Artificial Neural Network (ANN) for transient stability assessment (TSA) and an ANN-controlled Interline Power Flow Controller (IPFC) for transient stability enhancement (TSE) in TBHPP. The ANN was trained on 80 % of a 9828-sample dataset from different fault and load scenarios, with 20 % used for testing. For TSE, the ANN-controlled IPFC was evaluated under three-phase-to-ground faults, transmission line outages, and sudden load changes. For instance, in transient stability analysis under a three-phase-to-ground fault, the ANN-based IPFC achieves better performance than both the uncontrolled system and the PI-based IPFC, with significantly lower settling times and reduced overshoots. Specifically, the ANN-based IPFC reduces the settling time of rotor angle, rotor speed, output active power, and electromagnetic torque by 65.251 %, 65.190 %, 59.542 %, and 61.377 %, respectively, compared to an uncontrolled system, and by 59.358 %, 57.729 %, 51.382 %, and 54.771 %, respectively, compared to a PI-based IPFC system. Additionally, it decreases overshoot in rotor angle, rotor speed, output active power, and electromagnetic torque by 17.264 %, 41.667 %, 50.090 %, and 85.099 %, respectively, over an uncontrolled system, and by 7.914 %, 64.646 %, 42.798 %, and 36.885 %, respectively, over a PI-based IPFC system. This paper develops an ANN model for transient stability TSA and TSE. The TSA model applies linear and softmax activation functions, while TSE uses sigmoid. The ANN network has an input, 12-neuron hidden, and output layers. Results demonstrate ANN's effective stability assessment and show that ANN-based IPFC surpasses both PI-based IPFC and uncontrolled systems. The study was conducted using MATLAB/Simulink for simulation and script development.