Integrated Fault Detection Framework in Clean Energy Ecosystems: Analysis, Design, and Implementation for Advanced Generation, Storage, and Energy Management

Abstract

The rapid integration of clean energy ecosystems demands highly reliable fault diagnosis architectures to ensure operational safety. Traditional methods often struggle to decouple simultaneous faults in dynamic and noisy environments, leading to high missed detection rates or severe false alarms. To address this issue, this study proposes an Integrated Fault Detection Framework that synergizes the physical interpretability of analytical models with the high-dimensional feature extraction capabilities of deep learning. A hybrid architecture based on a Temporal Convolutional Network and Long Short-Term Memory (TCN-LSTM), constrained by equivalent physical models, is utilized to generate robust residual signals. Subsequently, Kernel Density Estimation (KDE) is applied to evaluate these residuals through non-parametric adaptive thresholding. Simulation results demonstrate that the proposed methodology achieves a diagnostic classification accuracy exceeding 98.8%, reduces the missed detection rate to approximately 20%, and completely eliminates false alarms even under severe environmental noise disturbances. Overall, this integrated approach provides a robust, real-time diagnostic solution to prevent catastrophic failures, such as battery thermal runaway, thereby ensuring the reliable operation of clean energy systems.