Using AI to leverage new forms of data in modelling cycling behaviours in the LCR

问题：据估计，英格兰有22%的成年人不运动，而利物浦城市地区（LCR）的每个地方政府（PHE 2019）的这一比例更高。体育活动是健康的重要决定因素，与降低心血管疾病风险以及改善心理健康有关。因此，设计鼓励身体活动的城市和社区是一项重要的政策重点。越来越多的人采用的方法是增加主动出行，特别是骑自行车。然而，在英国，只有3.3%的成年人每周至少骑3次自行车旅行，而LCR的比例更低。有针对性地投资自行车基础设施可以鼓励更多的人骑自行车，并减少空气污染，间接有益于健康。解决方案：该项目将使用最先进的机器学习和人工智能技术来利用新形式的数据来改善围绕自行车投资的决策。这种方法很少应用于交通建模，但为复杂（大）数据的处理和建模提供了新颖性，可以为骑行行为和基础设施提供信息。博士学位的一个关键优势将是定制方法的发展，能够最大限度地利用自行车背景下未开发的数据。这些方法的设计将使它们可以很容易地在任何地方政府中部署，以通知自行车供应。该项目旨在与非学术合作伙伴LCR共同开发现实世界的解决方案，从而最大限度地发挥影响力。博士学位大纲：该项目将以出版为基础的博士学位为结构，并将包括三个主要子项目：1。模拟充气道路管计数器的单车交通量本项目将利用充气道路管计数器的单车交通量，以及有关计数器所在位置特征的辅助数据，建立一个预测各段街道的单车交通量的模型，该模型可应用于整个路政铁路网络。这将加强对从公共卫生到交通规划等一系列领域相关机构中骑自行车者分布情况的了解。在方法上，该项目将扩展基于树木的模型（如随机森林、增强树），以明确地纳入空间特征和关系。理解循环流背后的驱动因素本文将揭示前一篇文章中获得的估计背后的驱动因素。通过将传统的社会经济数据来源（如人口普查、贫困得分）与识别环境特征（如道路质量、树叶等）的视频片段或图像数据等新方法相结合，该研究将确定环境因素如何与社会条件相互作用，以确定人们在不同地方骑自行车的程度。为了能够利用这些数据源，将需要最先进的人工智能技术，如卷积神经网络（cnn）。在这篇最后的论文中，学生将使用前两篇文章的结果来建立一个决策系统，为LCR改善自行车基础设施的政策提供信息。该系统将实现两个主要功能：首先，它将提供一种直观的方式，使所产生的预测模型的结果和与之相关的不确定性措施可视化和相互作用；其次，它将围绕基础设施的改善提出“假设”类型的问题。在此背景下，学生将探索空间交互和基于主体的模型的适用性。预期这一系统将能够确定长期制度内的政策优先事项。

The Problem:It is estimated that 22% of adults in England are physically inactive, and these rates are higher within each Local Authority within the Liverpool City Region (LCR) (PHE 2019). Physical activity is an important determinant of health, being associated with lower risk of cardiovascular diseases, as well as improved mental wellbeing. Designing cities and neighbourhoods to encourage physical activity is therefore an important policy priority. An increasingly adopted approach is to increase the uptake of active travel, particularly cycling. However, only 3.3% of adults in England cycle for travel at least 3 times per week, and rates are lower for the LCR. Targeted investment in cycling infrastructure can encourage more individuals to take up cycling, as well as reduce air pollution indirectly benefiting health.The Solution:This project will use state-of-the-art machine learning and AI techniques to leverage new forms of data to improve decision making around cycling investment. Such approaches are rarely applied within transport modelling, but offer novelty to process and model complex (big) data to inform cycling behaviours and infrastructure provision. A key advantage of the PhD will be the development of bespoke methods that enable to make the most out of data unexplored in the context of cycling. These methods will be designed so that they can be easily deployed within any local government to inform cycling provision. This project is designed to co-produce real-world solutions alongside the non-academic partner, the LCR, thus maximising impact.Outline of the PhD:The project will be structured as a publication-based PhD, and will include three main subprojects:1. Modelling volume of cycling traffic from pneumatic road tube countersThis project will use cycling counts from pneumatic road tube counters and ancillary data about the characteristics of the locations where they are placed to build a predictive model of cycling counts at the street segment that can be deployed to the entire network of the LCR. This will enhance the understanding of the distribution of cyclists to agencies related a range of domains, from public health to transport planning. Methodologically, this project will expand tree-based models (e.g. random forests, boosted trees) to explicitly incorporate spatial features and relationships.2. Understanding the drivers behind cycling flowsThis paper will unpack the driving factors behind the estimates obtained in the previous one. By combining traditional socio-economic sources of data (e.g. Census, Deprivation scores) with new approaches such as video footage or imagery data that recognise features of the environment (e.g. road quality, foliage, etc.), the study will identify how environmental factors interact with social conditions to determine the extent to which people cycle in different places. To be able to leverage these data sources, state-of-the-art AI techniques such as convolutional neural networks (CNNs) will be required.3. Predicting where to invest on urban cycling infrastructureIn this final paper, the student will use results from the previous two in order to build a decision-making system that informs policies on improvement of cycling infrastructure in the LCR. The system will fulfil two main functions: first, it will provide an intuitive way of visualising and interacting with the results of the predictive models generated and the measures of uncertainty associated with them; second, it will feature the capability of asking "what-if" type of questions around the improvement of infrastructure. In this context, the student will explore the suitability of spatial interaction and agent-based models. It is expected this system will enable the identification of policy priorities within the LCR.