Stacked neural filtering network for reliable NEV monitoring

Abstract

Reliable monitoring of new energy vehicles (NEVs) is crucial for ensuring traffic safety and energy efficiency. However, traditional Transformer-based methods struggle with quadratic computational complexity and sensitivity to noise due to the self-attention mechanism, leading to efficiency and accuracy limitations in real-time applications. To address these issues, we propose the Stacked Neural Filtering Network (SNFN), which replaces self-attention with a learnable filter block that operates in the frequency domain, reducing complexity to logarithmic-linear levels. This novel design improves computational efficiency, mitigates overfitting, and enhances noise robustness. Experimental evaluations on two real-world NEV datasets demonstrate that SNFN consistently achieves superior accuracy and efficiency compared to traditional methods, making it a reliable solution for real-time NEV monitoring.