Real Time Monitoring Method for Power Quality in Photovoltaic Substations under GA Optimized Convolutional Neural Network

This study proposes a real-time monitoring method based on GA optimized CNN for power quality monitoring in photovoltaic substations. Firstly, perform horizontal and vertical correction preprocessing on the power quality data, eliminate outliers and maintain time series characteristics, and then perform standardization processing. In the feature extraction stage, a CNN network architecture is adopted to extract perturbation features through convolutional layers, enhance nonlinear expression ability through ReLU activation function, reduce parameter size through pooling layer, and improve training stability through BN layer. Finally, perturbation classification is achieved through Softmax function. Further use genetic algorithm to optimize CNN hyperparameters, including dynamically adjusting crossover probability and mutation probability, evaluating network performance through fitness function, and iteratively optimizing to obtain the optimal network configuration. The experimental results show that the method proposed in this paper performs well in photovoltaic power quality monitoring, with a monitoring accuracy of 97.89% to 98.21% and a stable event response time of 0.78 to 0.95 seconds.