Scholarlink

<p>The digital twin system of sports venues often faces the problems of complex fusion of multi-source heterogeneous data and insufficient real-time performance. This article aims to develop an intelligent system that utilizes multimodal diffusion models and IoT data to address these challenges. The proposed framework integrates multimodal data, including video, sensor flow, and pedestrian flow counting. By adopting a cross modal attention mechanism for dynamic feature fusion and using a diffusion model for data augmentation and state prediction, the system has improved data reliability and prediction accuracy. The experiment conducted on a synthetic dataset consisting of 12,000 samples, validated by a real subset of local facilities, showed that the system achieved an accuracy of 95% in anomaly detection, which is 10% points higher than traditional methods. In the resource scheduling task, peak energy consumption was reduced by 25%, while inference time and median system response time were optimized to approximately 50ms. These results indicate that the proposed system effectively improves real-time performance and energy efficiency, providing a robust and scalable solution for intelligent operation of sports venues.</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

Research on the construction of a digital twin system for sports venues through the collaboration of multimodal diffusion models and Internet of Things data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月28日

查看全文
  • Xiaolei Xi

摘要

The digital twin system of sports venues often faces the problems of complex fusion of multi-source heterogeneous data and insufficient real-time performance. This article aims to develop an intelligent system that utilizes multimodal diffusion models and IoT data to address these challenges. The proposed framework integrates multimodal data, including video, sensor flow, and pedestrian flow counting. By adopting a cross modal attention mechanism for dynamic feature fusion and using a diffusion model for data augmentation and state prediction, the system has improved data reliability and prediction accuracy. The experiment conducted on a synthetic dataset consisting of 12,000 samples, validated by a real subset of local facilities, showed that the system achieved an accuracy of 95% in anomaly detection, which is 10% points higher than traditional methods. In the resource scheduling task, peak energy consumption was reduced by 25%, while inference time and median system response time were optimized to approximately 50ms. These results indicate that the proposed system effectively improves real-time performance and energy efficiency, providing a robust and scalable solution for intelligent operation of sports venues.