Scholarlink

<p>To address the problem of security analysis lag caused by the delay in dynamic topology updates in short video networks, this paper constructs a real-time digital twin framework based on edge computing lightweight proxies and achieves low-delay state response by accelerating the state synchronizer through field-programmable gate array (FPGA). For cross-modal attack detection blind spots, a joint analysis model is constructed by combining network topology and content features to improve the ability to identify complex threats and predict evolution trends. A lightweight proxy module is deployed on the content delivery network (CDN) node. Docker containerization technology is used to collect device states, user behavior, and traffic feature data in real time to build a local topology subgraph. A parallel state processor based on pipeline architecture. Incremental updates of the topological relationship matrix (500 × 500) are implemented, and the clock cycle is optimized through hardware description language. By integrating graph attention network (GAT) and bidirectional encoder representations from transformers (BERT) text encoders, a three-dimensional feature space of “node-edge-content” is established to detect topological anomalies and content violation coupled attacks of disguised accounts. A spatio-temporal graph convolutional network (STGCN) is constructed to extract topological evolution patterns in a fixed time window and predict the propagation paths of high-risk nodes. Experimental results show that the synchronization delay of FPGA tends to 0.5ms in the topology synchronization change time of 100ms, which has a high response speed; in the four types of cross-modal coordinated attack scenarios, the average identification precision and recall of the joint analysis model in this paper are 91.9%±0.9% and 88.7%±1.4%, respectively, and its cross-modal attack identification performance is strong; in the prediction of node connection probability, the root mean squared error (RMSE) of STGCN is 6.84 ± 0.31 at a time offset of 2.5&#xa0;s, which verifies the model’s ability to predict evolution trends; the total end-to-end delay is 7.5ms. In the 72-hour continuous stress test, the central processing unit (CPU) usage rate is 8.3%→9.1%; the memory leak rate is &lt; 0.01%/h; the false positive rate for abnormal events is 0.7%→1.2%; the prediction error accumulation rate is 0.05%/h, verifying the long-term prediction stability. The data results demonstrate the effectiveness of this paper’s research on secure propagation analysis of short videos and evolution trend prediction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Topology analysis and evolution trend prediction of secure propagation networks for short videos based on digital twin model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月21日

查看全文
  • Yu Wang,
  • Wei Liu

摘要

To address the problem of security analysis lag caused by the delay in dynamic topology updates in short video networks, this paper constructs a real-time digital twin framework based on edge computing lightweight proxies and achieves low-delay state response by accelerating the state synchronizer through field-programmable gate array (FPGA). For cross-modal attack detection blind spots, a joint analysis model is constructed by combining network topology and content features to improve the ability to identify complex threats and predict evolution trends. A lightweight proxy module is deployed on the content delivery network (CDN) node. Docker containerization technology is used to collect device states, user behavior, and traffic feature data in real time to build a local topology subgraph. A parallel state processor based on pipeline architecture. Incremental updates of the topological relationship matrix (500 × 500) are implemented, and the clock cycle is optimized through hardware description language. By integrating graph attention network (GAT) and bidirectional encoder representations from transformers (BERT) text encoders, a three-dimensional feature space of “node-edge-content” is established to detect topological anomalies and content violation coupled attacks of disguised accounts. A spatio-temporal graph convolutional network (STGCN) is constructed to extract topological evolution patterns in a fixed time window and predict the propagation paths of high-risk nodes. Experimental results show that the synchronization delay of FPGA tends to 0.5ms in the topology synchronization change time of 100ms, which has a high response speed; in the four types of cross-modal coordinated attack scenarios, the average identification precision and recall of the joint analysis model in this paper are 91.9%±0.9% and 88.7%±1.4%, respectively, and its cross-modal attack identification performance is strong; in the prediction of node connection probability, the root mean squared error (RMSE) of STGCN is 6.84 ± 0.31 at a time offset of 2.5 s, which verifies the model’s ability to predict evolution trends; the total end-to-end delay is 7.5ms. In the 72-hour continuous stress test, the central processing unit (CPU) usage rate is 8.3%→9.1%; the memory leak rate is < 0.01%/h; the false positive rate for abnormal events is 0.7%→1.2%; the prediction error accumulation rate is 0.05%/h, verifying the long-term prediction stability. The data results demonstrate the effectiveness of this paper’s research on secure propagation analysis of short videos and evolution trend prediction.