Performance Validation of a Deep Learning Framework for Power Quality Disturbance Classification in Renewable Energy-Based Smart Grids

Abstract

The expansion of solar PV systems and wind energy and battery storage, and the integration of power-electronic converters in smart control grids, offers a new set of issues regarding the identification of power quality (PQ) and PQ disturbances (PQDs). Deep learning methods can be applied since they can automatically learn complicated temporal, spectral and waveform characteristics that are difficult to build or model from predefined features. For the classification of multi-class disturbances, deep learning models based on advanced architectures such as Convolutional Neural Networks (CNN), Long Short Term Memory (LSTM) networks, and hybrid deep learning models are used in conjunction with signal processing techniques such as wavelet or temporal frequency transformation. Analyse our framework in terms of computational speed, robustness, and qualitative performance based on accuracy, recall, F1-score and study of confusion matrix. The suggested predictor framework has a significant explanatory power in an exemplary regression model with R = 0.835 and R2 = 0.697. The statistics below are only examples, as no real data on individual surveyed was available. Instead, it calls for verification of various parameters like quantification against diverse noise categories and other operations such as load variations, compounded disturbances, and evaluations using unseen test schemes. The proposed framework provides a method-driven basis for systematic identification of PQDs to allow direct measurements, greater grid awareness, and increased dependability in incorporating new renewable power within future smart-grid platforms.