GA-optimized Deep Neural Network for intelligent power-quality regulation in power systems

Abstract

To address power-quality deterioration in increasingly complex power-system operating conditions, this paper develops a genetic-algorithm-optimized deep neural network (GA-DNN) for intelligent regulation. The GA module searches the DNN hyperparameter space, including layer depth, hidden-unit width, learning rate, batch size, dropout rate, activation function, and training epoch number, while the DNN learns nonlinear mappings among voltage amplitude, harmonic content, frequency, phase angle, and disturbance type. Verification based on the Power Quality Dataset shows that the proposed model reduces voltage THD from 6.8% to 3.2%, decreases frequency deviation from 0.14 Hz to 0.04 Hz, shortens average response delay from 65 ms to 28 ms, and improves recognition accuracy from 84.5% to 95.8%. These results indicate that GA-DNN can strengthen adaptive power-quality monitoring and regulation under voltage fluctuation, harmonic pollution, and frequency instability scenarios.