A novel deep learning-based control for voltage sag prediction and DVR–LVRT coordination in grid-connected wind turbine systems

This study proposes a novel deep feedforward neural network (DFNN)-based control for voltage sag prediction and dynamic voltage restorer (DVR) integration with low-voltage ride-through (LVRT) to improve voltage stability and energy efficiency in wind turbine systems. Proposed approach precisely predicts sag duration, adaptively regulates DVR activation according to LVRT profile to avoid redundant compensation. This predictive control reduces DVR operating time and energy consumption while maintaining voltage stability. Synthetic datasets representing various grid conditions, including distorted voltage, enable DFNN to achieve a prediction accuracy near 99 %, a time error margin of 0.005–0.02 s and a response time of 0.016 s. This precise prediction improves DVR energy efficiency by nearly 30 % per long-duration fault. Compared to a support vector regression (SVR) model, DFFN achieves 33.3 % faster response and lower error metrics. Simulation results validated in a MATLAB/Simulink demonstrate effectiveness of proposed approach in enhancing LVRT capability and overall grid efficiency.