EAGER: Feedback-based Network Optimization for Smart Cities

EARLY概念探索性研究资助（EAGER）项目将专注于未来智慧城市中的人类与基础设施的互动。具体来说，该项目的目标是开发一个概念性的网络优化框架，利用来自众包数据的用户反馈。下一代无处不在的交通传感器，例如固定交通检测器、具有基于位置的服务的移动的传感器（例如，Google流量）和流量主动移动的传感器（例如，Waze）已经被用于检索大量的和不同的地理定位和时间戳数据。此外，来自社交媒体和移动的应用程序的众包数据越来越多地用于了解人类移动性。另一方面，现有的交通网络优化模型在很大程度上没有考虑人的行为。该项目将探索设计未来电动汽车无线充电网络的方法。这样的充电网络将提供特别适合城市环境的不间断、动态充电。对于许多用户来说，无线充电将是一种机会主义的紧急充电选择，可以补充家庭、工作场所和零售设施的分布式充电资源。传统的网络优化模型隐含地假设充电需求是分布式的并且是给定的，在诸如找到包含无线充电网络段的最佳路线之类的任务中可能不会导致用户满意的解决方案。PI旨在通过开发：（i）明确考虑用户满意度的用户反馈驱动的网络优化模型，以及（ii）快速和可扩展的优化方法，用于有效利用众包信息的真实世界，大规模问题实现。这项研究的结果将包括一个“生活实验室”评估，学生将开发一个移动的应用程序，收集有关无线充电网段路线的反馈。这种反馈将被分析并纳入优化模型。这种方法将使用大规模机器学习和优化求解器的组合来实现。该移动的应用程序将通过安全渠道提供有关用户、行程类型和路线反馈的基本匿名信息。这些数据和所有路段的公开信息将用于探索相关的潜在因素，并创建基于成本敏感的支持向量机分类器，以确定最适合无线充电网络的路段，并调整基于优化的网络设计。PI将采用基于代数多重网格的方法来科普大规模的现实问题，而不是使用传统的、计算昂贵的优化求解器，并利用他们的支持向量机求解器。

This EArly-concept Grant for Exploratory Research (EAGER) project will focus on human-infrastructure interactions within future smart cities. Specifically, the goal of this project is to develop a conceptual network optimization framework that exploits user feedback from crowd-sourced data. The next-generation ubiquitous traffic sensors, such as fixed traffic detectors, mobile sensors with location-based services (e.g., Google traffic), and traffic active mobile sensors (e.g., Waze), have been used to retrieve large and diverse geo-located and time-stamped data. In addition, data crowd-sourced from social media and mobile applications are increasingly available for understanding human mobility. On the other hand, existing network optimization models in transportation were largely developed without considering human behavior. This project will explore methods to design an electric vehicle wireless charging network of the future. Such a charging network will provide non-stop, in-motion charging particularly suitable for urban environments. For many users, wireless charging will be an opportunistic, emergency charging choice that supplements distributed charging resources at home, the workplace, and at retail facilities. Traditional network optimization models, which implicitly assume that charging demand is distributed and a given, may not result in a user-satisfactory solution in such tasks as finding the best routes that contain wireless charging network segments. The PIs aim to overcome this deficiency by developing: (i) user feedback-driven network optimization models that explicitly account for user satisfaction, and (ii) fast and scalable optimization methods for real-world, large-scale problem implementations that efficiently and effectively utilize crowd-sourced information. The results of this research will include a "living lab" assessment in which students will develop a mobile app to collect feedback about the route with wireless charging network segments. This feedback will be analyzed and incorporated in optimization models.This approach will be implemented using a combination of large-scale machine learning and optimization solvers. The mobile app will provide basic anonymized information about users, trip types, and route feedback through secured channels. These data and open information about all road segments will be used to explore relevant latent factors and create cost-sensitive support vector machine based classifiers to identify the most suitable segments for wireless charging network and adjust the optimization-based network design. Rather than using traditional, computationally expensive optimization solvers, the PIs will pursue an algebraic multigrid-based approach to cope with large-scale, real-world problems, and leverage their support vector machine solvers.