可解释AI驱动的WSN智能检测方法OA

To address the need for efficient,lightweight,and interpretable detection,E²HEN,an intrusion detection framework that integrates explainable learning with temporal reconstruction was proposed.The framework employed an explainable boosting machine(EBM)in conjunction with SHAP(SHapley Additive exPlanations)to generate interpretable feature attribution vectors,enhancing decision transparency.An attention-enhanced bidirectional autoencoder(BAED)was further utilized to perform fine-grained reconstruction of abnormal sequences.Experiments conducted on a hybrid dataset combining NSL-KDD and WSN-DS demonstrate that E²HEN achieves superior performance over baseline models in terms of F1-score,accuracy,and interpretability.Specifically,it reaches 98.9%detection accuracy for typical attacks and an overall F1-score of 97.6%.In addition,the system supports dynamic threshold adjustment based on node residual energy,achieving a balance between detection performance and energy efficiency.The study also incorporates multi-perspective visualization to explain model decisions,offering a reliable and deployable approach for WSN security monitoring.