The Industrial Internet of Things (IIoT), an emerging technology with significant potential, has been widely adopted in industrial production. However, IIoT faces increasingly severe threats from malicious attacks. To effectively detect and predict malicious attacks, Network Security Situation Prediction (NSSP) technology has become a focal point of research in academia and industry. Nevertheless, the existing deep-learning-based NSSP solutions often neglect the interconnectedness and temporal characteristics of data, thereby limiting their performance enhancement. To overcome this limitation, this paper proposes a novel NSSP method based on the Linear Gated Future Prediction Network (LGF-Net). Specifically, the proposed method integrates the strengths of U-Net and ResNet, significantly enhancing the performance of NSSP. The LGF-Net model enhances traditional deep learning networks in two key aspects: firstly, it effectively captures the inherent correlations in time series through a multi-level structure; secondly, it introduces an information compensation mechanism that significantly reduces information loss during the learning process. These improvements substantially enhance the model’s ability to capture correlations between elements, enabling it to accurately forecast future network security situations using multi-level historical data. Furthermore, we evaluate the computational efficiency of the proposed approach through theoretical complexity analysis. Finally, extensive experimental results demonstrate that the proposed method exhibits significant advantages in multiple performance metrics compared to existing approaches.