Speaker
摘要
烟气控制系统是保障基础设施防火安全的关键组成部分。传统烟气控制系统设计主要依赖逐案分析的范式，决策基于有限设计方案之间的比较，缺乏针对多样化基础设施构型与火灾场景的数据驱动优化方法。为克服这些挑战，本研究提出了一种融合扩散模型与NSGA-II的AI驱动生成式设计框架，用于优化烟气控制系统的方案设计。考虑基础设施构型的多样性与火灾场景的不确定性，构建了扩散模型以高效预测烟气控制性能。同时，基于扩散模型构建了融入火灾风险评估指标的NSGA-II框架，通过对设计空间的全局探索，能够自动生成面向随机场景的风险-成本最优方案。以隧道火灾场景为例，该方法对训练数据集内烟气场的预测准确率达到95%，对训练分布之外的隧道几何构型及火灾场景的预测准确率仍保持在80%以上。针对纵向通风、集中排烟及其组合这三种主导烟气控制模式，优化方案可在数分钟内自动生成，与传统方法相比，通风量分别降低52%、15%和23%，火灾风险分别降低10%、6%和12%。本研究提出了一种稳健、可泛化且面向风险最小化的决策工具，为消防工程中智能化烟气控制系统的研发提供了支持。
Abstract
Smoke control systems are a critical component for ensuring the fire safety of infrastructure. Traditional smoke control system design predominantly relies on a case-by-case paradigm, with decision-making based on comparisons among a limited set of design schemes, lacking data-driven optimization methods that can accommodate diverse infrastructure configurations and fire scenarios. To overcome these challenges, this study proposes an AI-driven generative design framework that integrates a diffusion model with NSGA-II to optimize smoke control system schemes. Considering the diversity of infrastructure configurations and the uncertainty of fire scenarios, a diffusion model is developed to efficiently predict smoke control performance. Meanwhile, a diffusion model-based NSGA-II framework is constructed by incorporating fire risk assessment indicators, enabling automatic generation of risk-cost optimal solutions for random scenarios through global exploration of the design space. Taking tunnel fire scenarios as a case study, the method achieves 95% prediction accuracy of smoke fields for the training dataset and maintains over 80% accuracy for tunnel geometries and fire scenarios outside the training distribution. For the three dominant smoke control modes—longitudinal ventilation, centralized smoke exhaust, and their combination—optimized schemes can be automatically generated within minutes, reducing ventilation rates by 52%, 15%, and 23%, and fire risk by 10%, 6%, and 12%, respectively, compared with traditional methods. This study presents a robust, generalizable, and risk-minimization-oriented decision-making tool, supporting the development of intelligent smoke control systems in fire protection engineering.
| 关键词 | AI驱动生成式设计，烟气控制系统，扩散模型，NSGA-II，数据驱动优化，隧道火灾 |
|---|---|
| Keywords | AI-driven generative design, smoke control system, diffusion model, NSGA-II, data-driven optimization, tunnel fire |
