Multi scale attention driven DACDiff+ distributed power FDIAs defense model

This paper proposes a novel defense model called DACDiff + to address false data injection attacks (FDIAs) in distributed power systems. This model integrates multi-scale attention mechanism and dynamic adaptive diffusion model, which can effectively capture short, medium, and long-term dependencies in time series data and dynamically adjust the denoising step size according to the attack intensity. The experimental results show that DACDiff+ improves data recovery accuracy by 32.7% compared to traditional statistical methods, achieves a defense success rate of 86.7% against stealth attacks, and has a inference time of 32.5ms, meeting real-time requirements. The model validated the key role of multi-scale attention mechanism and dynamic diffusion model through ablation experiments, providing an efficient solution for smart grid security.