Active network cyber threat detection and risk mitigation for intelligent driving combining data dimensionality reduction and CNN-LSTM

With the continuous advancement of intelligent driving technology, automotive communication networks face increasingly complex threats from advanced cyber attacks. Traditional intrusion detection methods struggle to simultaneously meet the requirements of high accuracy, low latency, and lightweight deployment. To address these problems, this study first designs a local-global joint adaptive projection dimensionality reduction method. Second, this study constructs a lightweight Convolutional Neural Network-Long Short-Term Memory model based on Depthwise Separable Convolution. This model introduces a feature reconstruction verification layer and a joint loss function to enhance feature preservation capability. Finally, this study establishes a dimensionality reduction-driven integrated active risk mitigation mechanism to achieve threat hierarchical assessment, explainable traceability, and adaptive closed-loop response. Experimental results demonstrate that this model achieves an accuracy of 0.982, a recall of 0.979, and an F1 score of 0.980. The average detection latency is 5.2 ms for HCRL and 6.8 ms for VeReMi. The parameter size is only 1.28 M for HCRL and 1.51 M for VeReMi. Under denial-of-service attacks, the hierarchical response accuracy of the active mitigation mechanism only decreases by 5.7% as the attack frequency increases, which is much lower than that of the baseline models. Experimental evaluation demonstrated that the proposed method achieved lower computational overhead and response latency while maintaining high detection accuracy. Meanwhile, it realizes explainable localization of threat sources and active closed-loop control. This study provides an efficient, deployable active cybersecurity solution for resource-constrained in-vehicle environments.