A Glass-Box AI Framework for Predicting and Explaining Urban Solar Energy Generation

Yumin Zhou, Whye Loon Tung, Hiok Quek

Keywords: Urban AI, Explainable AI (XAI), Trustworthy AI, Solar Forecasting, Neuro-Fuzzy Systems

Abstract: The reliable integration of volatile solar power is critical for sustainable urban energy systems. Engineers face a choice between transparent classical models that lack accuracy and high-performing black-box' models that pose operational risks. This paper breaks that trade-off by introducing L-FMLC, a 'glass-box' AI framework that delivers both state-of-the-art performance and deep interpretability. L-FMLC provides instance-specific linear equations for every prediction, grounded in autonomously discovered fuzzy sets that represent intuitive concepts like 'clear' and 'overcast' skies. It then distills thousands of potential decision paths into a handful of strategic 'meta-rules', culminating in natural-language reports. In a real-world solar forecasting case study, we show L-FMLC surpasses classical and standard deep learning baselines while providing this full stack of actionable explanations. This work offers a practical blueprint for building AI systems for urban infrastructure that are simultaneously high-performing and fundamentally trustworthy.

Submission Number: 7