EfficientNetB3-Based Framework for Real-Time Disturbance Classification in Smart Grid Systems Using PMU Data

This paper proposes a deep learning-based framework using the EfficientNetB3 architecture to classify disturbances in smart grid systems based on Phasor Measurement Unit (PMU) data. The PMU data, which captures voltage and frequency measurements, is transformed into pseudo-color images representing faults, loss of generation, and synchronous motor switching events. These images were used to train the EfficientNetB3 model, which obtained a stunning accuracy level of 93% on the training set. The model provided a good accuracy of 78% on the test set, with almost perfect precision for fault detected (0.99) and fair precision and recall of 0.82 and 0.88, respectively, for the loss of generation. However, the model was tried and tested, and there were issues in handling less frequent occurrences than with synchronous motor switching because of the lack of enough data. The results indicate that using the EfficientNetB3 model might increase the efficiency of smart grid monitoring and protection systems in identifying and classifying grid disturbances.