Real-Time Simulation of Urban Systems Using Deep Learning

使用深度学习实时模拟城市系统

• 批准号: 2749683
• 金额: --
• 依托单位: Newcastle University
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2022
• 资助国家: 英国
• 起止时间: 2022 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2749683
• 关键词: Real; Time; Simulation; Urban; Systems

My research aims to address some of the complex challenges inherent in modelling urban systems, which exhibit non-deterministic and non-linear behaviour's. It highlights the limitations of existing solutions, particularly agent-based models (ABMs), due to the absence of robust validation methods, and inability to handle real-time data. My project will take employ a hybrid deep-learning model to create an 'emulator' of an ABM, bridging the gap between theoretical models and real data. By leveraging existing centralised repositories of real-time urban data, the research aims to predict the spatio-temporal dynamics of urban agents. This approach has the potential to move us closer to the creation of an 'urban digital twin,' offering insights into the behaviour of complex urban systems and non-independent sensors' spatio-temporal dependencies, allowing for more effective response to developing events, like overcrowding or flooding.

我的研究旨在解决城市系统建模中固有的一些复杂挑战，这些挑战表现出不确定性和非线性行为。它突出了现有解决方案的局限性，特别是基于代理的模型（ABMs），由于缺乏健壮的验证方法，并且无法处理实时数据。我的项目将采用混合深度学习模型来创建一个ABM的“模拟器”，弥合理论模型和实际数据之间的差距。通过利用现有的实时城市数据集中存储库，该研究旨在预测城市代理的时空动态。这种方法有可能使我们更接近于创建“城市数字双胞胎”，提供对复杂城市系统行为和非独立传感器时空依赖关系的见解，允许更有效地应对发展中的事件，如过度拥挤或洪水。