Temporal-Convolutional Adversarial Autoencoding With Channel-Wise Attention for Intrusion Detection in the Internet of Vehicles

The Internet of Vehicles (IoV) lets cars talk to one another in smart ways by sharing data in real time between vehicles, infrastructure, and edge nodes. But as more and more parts become linked, the danger of advanced attacks rises. At the same time, traditional Intrusion Detection Systems (IDS) typically cannot keep up with IoV's changing, low-latency, and resource-limited needs. Deep learning has become a viable alternative, but it needs a lot of labeled data and architectures that are hard to compute, which makes it less useful in real-world vehicle situations. To solve this problem, we suggest Vehicular Intrusion Detection by Temporal-aware Attention (VITA), a lightweight, self-supervised, and label-free IDS framework made just for IoV. VITA uses an Adversarial Autoencoder (AAE) with Efficient Channel Attention (ECA) to pick out important spatial characteristics and residual Temporal Convolutional Networks (TCNs) to simulate long-range temporal relationships in a way that is efficient way. A latent-space smoothing technique is added to stabilize adversarial learning, and a log-cosh reconstruction loss is added to make the system more resistant to noisy vehicle telemetry. VITA works in real time and does not need a lot of processing power, so it can be used on edge devices in vehicles. Our proposed VITA framework shows superior performance across several critical criteria, including an average 30% reduction in latency compared to state-of-the-art models. VITA achieves an average 8% improvement in detection accuracy, effectively identifying a broad spectrum of vehicular attacks, such as replay, Global Positioning System (GPS) spoofing, and Denial-of-Service (DoS) attacks. Furthermore, VITA outperforms existing systems with an average 5x reduction in false positive rate, and shows robust performance under varying noise levels, with only a modest 4.7% drop in accuracy under high jitter conditions. Additionally, it excels in computational efficiency, requiring 50% less memory and processing power on average, making it highly suitable for real-time, edge-based IoV deployments.