As the Industrial Internet of Things becomes increasingly interconnected with critical infrastructure, intrusion traffic exhibits characteristics such as high-dimensional redundancy, class imbalance, and temporal correlation, posing challenges for detection systems in terms of feature representation, model complexity control, and real-time performance. To address the aforementioned issues, this paper proposes an intrusion detection algorithm based on feature optimization and asynchronous advantage actor-critic learning (APA3CID). First, the raw dataset was preprocessed using methods such as label encoding and normalization. Feature selection was performed using the improved Whale Optimization Algorithm (WOA) to reduce data redundancy and eliminate irrelevant features. The samples were then serialized based on the order in which they were collected. Second, we model the detection process as a Markov decision process, use a sliding window to construct states that capture recent temporal features, and, building upon the Asynchronous Advantage Actor-Critic (A3C) framework, we incorporate an adaptive exploration mechanism to address the issues of insufficient exploration in the early training phase and unstable convergence in the later phase. Additionally, we introduce an asynchronous lag correction strategy that utilizes truncated importance weights to mitigate the bias caused by policy lag in asynchronous parallel training, thereby enhancing the stability and robustness of policy updates. Finally, experimental results show that on the X-IIoTID dataset, APA3CID achieves a 3.51% increase in detection rate and a 4.26% increase in F1-score compared to the traditional A3C algorithm. On the WUSTL-IIoT-2021 dataset, single-sample prediction takes as little as 11.56 microseconds, with Acc, DR, and F1-score all exceeding 90%. This outperforms comparison models such as LR, XGBoost, CNN, and the baseline A3C, meeting the requirements of industrial IoT scenarios for low false-negative rates and high real-time performance.

The proliferation of next-generation Internet infrastructures and the Internet of Things (IoT) has exponentially increased network traffic complexity. While deep learning (DL)-based intrusion detection systems (IDSs) show immense potential, they persistently suffer from challenges including high computational overhead, vanishing gradients in deep architectures, and acute sensitivity to noise. Consequently, these issues impede their real-time deployment in resource-constrained edge computing environments. To overcome these limitations, we propose a novel, lightweight, and robust intrusion detection framework based on deep neural networks (DNNs). Initially, we employ a Robust Scaler-based statistical preprocessing strategy to supersede traditional Z-score standardization, effectively mitigating the adverse impacts of outliers and burst traffic noise. Subsequently, we design an advanced architecture that integrates self-normalizing residual blocks with a channel attention mechanism. Leveraging compressed hidden layers alongside the Scaled Exponential Linear Unit (SELU) activation function, this architecture not only mitigates the vanishing gradient problem but also amplifies critical traffic features. Concurrently, it achieves a substantial reduction in both parameter count and inference latency. Furthermore, we introduce a cosine annealing strategy to dynamically adjust the learning rate during training, thereby facilitating the model’s escape from local optima and accelerating convergence. Extensive experiments on standard benchmark datasets demonstrate that our proposed framework achieves superior detection accuracy while maintaining exceptional computational efficiency compared to state-of-the-art baselines.