F2GAN: A Feature-Feedback Generative Framework for Reliable AI-Based Fault Diagnosis in Inverter-Dominated Microgrids

TL;DR

The paper tackles the challenge of data scarcity and class imbalance in AI-based fault diagnosis for inverter-dominated microgrids by introducing F2GAN, a feature-feedback GAN that enforces mean-variance alignment and correlation preservation in the generator. It develops a three-phase framework combining microgrid modeling, synthetic data augmentation, and real-time validation, and demonstrates superior distributional fidelity and downstream diagnostic performance against CGAN, WGAN-GP, and TVAE using TSTR evaluations. The method leverages a statistically regularized generator loss $L_G = L_{adv} + \lambda_{MV} L_{MV} + \lambda_{corr} L_{corr}$ and a feedback mechanism to maintain inter-feature dependencies, achieving near-perfect real-time fault classification on a hardware-in-the-loop platform. The results indicate that synthetic fault data generated by F2GAN can bridge the gap between simulated and real-world microgrid fault datasets, enabling reliable, data-efficient fault diagnosis and potential deployment in practical intelligent grid monitoring systems.

Abstract

Enhancing the reliability of AI based fault diagnosis in inverter dominated microgrids requires diverse and statistically balanced datasets. However, the scarcity and imbalance of high fidelity fault data, especially for rare inverter malfunctions and extreme external line faults, limit dependable model training and validation. This paper introduces a unified framework that models a detailed inverter dominated microgrid and systematically generates multiple internal and external fault scenarios to mitigate data scarcity and class imbalance. An enhanced generative model called F2GAN (Feature Feedback GAN) is developed to synthesize high dimensional tabular fault data with improved realism and statistical alignment. Unlike conventional GANs, F2GAN integrates multi level feedback based on mean variance, correlation, and feature matching losses, enabling the generator to refine output distributions toward real fault feature spaces. The generated datasets are evaluated through quantitative and qualitative analyses. Train on Synthetic, Test on Real (TSTR) experiments demonstrate strong generalization of machine learning classifiers trained exclusively on F2GAN samples. The framework is validated on a hardware-in-the-loop (HIL) fault diagnosis platform integrated with a real time simulator and graphical interface, achieving 100 % diagnostic accuracy under real-time testing. Results confirm that F2GAN effectively bridges the gap between simulated and real world microgrid fault datasets

Figures (6)

• Figure 1: Architecture for Fault Data Generation and Detection: This framework illustrates the generation of simulation data from a real-time simulator (Phase 1), the traditional GAN-based architecture for synthesizing internal and external fault data (Phase 2), and the training of a fault detection model to classify fault scenarios accurately (Phase 3).
• Figure 2: Microgrid architecture and fault classification illustrating the interconnected MV and LV clusters of the microgrid, featuring external faults (LG, LLG, LL and LLL) and internal faults within the inverter's physical layer.
• Figure 3: Proposed F2GAN architecture with a feedback module providing mean–variance and correlation feedback to enhance generator learning.
• Figure 4: KDE plots comparing real and synthetic distributions for voltage–current pairs in phase-a line sections 12 and 23 under external faults. F2GAN displays strong density alignment with real data.
• Figure 5: TSTR evaluation metrics across generative models (CGAN, WGAN-GP, TVAE, F2GAN).