Automatic classification of Power Quality Disturbances (PQDs) is a challenging concern for both the utility and industry. In this paper, a novel technique for automatic classification of single and hybrid PQDs is proposed. The proposed algorithm consists of the Discrete Wavelet Transform (DWT) and Probabilistic Neural Network based BAT (PNN-BAT) optimal feature selection of PQDs. DWT with Multi-Resolution Analysis (MRA) is used for the feature extraction of the disturbances. The power quality disturbances are in the form of signals like voltage sag, voltage swell, voltage transients, flicker, voltage imbalance, and harmonics. Such disturbance signals cover a broad frequency spectrum because of its high sampling rate and produce megabytes of data which leads to the requirement of high storage space. In this paper, discrete wavelet transform is used to analyze the power quality disturbance signals and to reduce the storage space required. The PNN classifier is used as an effective classifier for the classification of the PQDs. However, the two critical concerns such as the selection of the optimal features and the spread constant value might affect the performance of the classifier. Hence, these two issues are addressed using a novel technique PNN-BAT based optimal feature selection and parameter optimization for improving the performance of the classification system. The BAT algorithm is used to select optimal features from a large feature set and the optimal value of the PNN spread constantly. The optimal feature selection method retains the useful features and discards the redundant features. [ABSTRACT FROM AUTHOR]