Work

MODWT-driven explainable spatio-temporal IDS for securing EV charging infrastructures

Year: 2026
Type: article
Abstract: The rapid deployment of Electric Vehicle Charging Stations (EVCS) has introduced significant cybersecurity challenges, as these infrastructures are increasingly targeted by adversaries seeking to disrupt charging services, manipulate billing, or compromise user privacy. Conventional intrusion detection systems often fail to detect attacks due to the dynamic and non-linear nature of EVCS power consumption data. To address this, we propose a Host-based Intrusion Detection System (HIDS) integrating Maximum Overlap Discrete Wavelet Transform (MODWT) feature extraction with deep learning architectures – LSTM, GRU, Transformer, and ModWaveMLP- to enable multi-resolution analysis of power consumption signals, effectively decomposing consumption patterns to reveal subtle anomalies overlooked in raw time-series data. Explainable AI techniques such as SHAP and LIME are incorporated to interpret model decisions with quantified stability, fidelity, and sparsity metrics. Experimental evaluation on the CICEVSE2024 dataset demonstrates that the GRU combined with db1 wavelet decomposition achieves the highest AUC of 0.9937, representing a 38.9% point improvement over baseline models without wavelet preprocessing. A comprehensive ablation study confirms SMOTE as the optimal class imbalance strategy, achieving 0.9945 AUC compared to 0.9738 for class-weighted training, establishing a transparent, wavelet-driven IDS framework tailored for next-generation EV charging infrastructure security.
Language: English
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7225894806
FWCI:
Cites: 32
Cited by:
Related to: 10