Fog-enabled deep learning framework for power quality enhancement in smart distribution systems

This paper introduces a novel fog-enabled deep learning (DL-Fog) framework for power quality (PQ) monitoring and mitigation in electrical distribution systems. Unlike traditional centralized methods, which are hindered by high latency and scalability issues, the proposed architecture integrates convolutional and recurrent neural networks with fog computing nodes deployed at the network edge. This enables low-latency, high-accuracy detection, classification, and prediction of PQ disturbances—including voltage sags/swells, flicker, and harmonic distortion—directly from raw waveform data. The key contributions of this work include: (i) a rigorous mathematical formulation of PQ phenomena using power system equations and harmonic models; (ii) a hybrid CNN–LSTM architecture that fuses time-domain and frequency-domain features for robust event recognition; (iii) a fog computing deployment strategy that optimizes inference latency and resource allocation under constraints of compute, memory, bandwidth, and quality-of-service (QoS); and (iv) an integrated optimization framework combining PQ detection with corrective control dispatch through inverter-based resources and dynamic filters. Comprehensive simulation results, including scalability tests on IEEE 69-bus and 119-bus feeders and robustness analysis under stressful conditions, demonstrate the effectiveness of the proposed approach, achieving over 98 % event classification accuracy, a 64.7 % reduction in voltage deviation, a 40.3 % reduction in total harmonic distortion, and sub-50 ms end-to-end response latency. These results affirm the framework's suitability as a comprehensive and integrated solution for scalable, intelligent, and responsive PQ management in future smart grids.