Power quality is main concern for the electrical energy consumptions and electrical equipment. Hence, the power quality disturbances needed to monitor, improve and control. However, most of the research are focusing to the accuracy of the classification analysis. In this paper, an approach to classify the power quality disturbances is presented using the deep neural network algorithm. A raw data containing various types of the power quality disturbances, like swell, interruption, harmonics, and normal signal is evaluated. This several types of power quality disturbance will be extracted using the Sparse Autoencoder (SAE). The various values of weight decay parameter, $\lambda$ and sparsity parameter, $\rho$ are applied to determine which features give optimal values. Optimal features learned from the SAE are then used to train a neural network classifier for identifying power quality disturbances.