Deep power: Deep learning architectures for power quality disturbances classification

The transformation of the conventional electric power grid to modern smart grid are subjected to power system quality and reliability problems. In order to ensure reliable, secure and quality supply of power, it is important to characterize and classify the power quality disturbances. Power quality (PQ) disturbance classification schemes implicitly relies o n feature engineering to extract unique and accurate features such as statistical information, spatio-temporal characteristics, stationary and non-stationary behavior of PQ signals. This paper explores the potentiality of deep learning algorithms to characterize and classify various PQ disturbances in smart grid. Deep learning algorithms have the inherent capability to automatically learn optimal features from raw input data and thus to avoid time-consuming feature engineering. To understand the effectiveness of various deep learning mechanisms, different architectures namely convolution neural network (CNN), recurrent neural network (RNN), identity-recurrent neural network (I-RNN), long short-term memory (LSTM), gated recurrent units (GRU) and convolutional neural network-long short-term memory (CNN-LSTM) are studied in this paper. Several experiments are conducted to propose an optimal deep learning architecture with specific network parameters and topologies. The performance of the proposed deep learning architecture is evaluated on a set of synthetic single and combined PQ disturbances and real-time PQ events. The proposed architecture is found to be accurate for real-time characterization and classification of power quality disturbances in smart grid.