摘要: |
This project aims to help transport agencies use “big data” to help mitigate congestion and manage system performance, for both freeways and arterials. Two investigatory objectives are planned. In the first, the research team will create and train an algorithm to spot the onset of incidents and recurring congestion, so that system managers can be more responsive. The hypothesis is that early responses help reduce the impacts (the queues are shorter, disappear quicker, and create less delay). The team will (1) fuse data such as real time traditional detector data, CV data, and other online data, to produce a significant and consistent data-stream of high volume and high velocity heterogeneous data; and (2) use deep reinforcement learning to train an AI-based algorithm to spot the onset of these events, distinguish between them, and generate response suggestions based on effective past system responses or modeling of the system with selected strategies. In the second investigatory thread, the team will create a performance monitoring algorithm that uses policy-based targets (e.g., speeds of 45 mph or better during congested conditions) and “big data” technologies to help agencies improve the efficacy of their congestion mitigation efforts. The team will use condition-based policy travel rates to simplify the inference process and ensure that performance management focuses on situations that have the greatest need. In both these efforts, the team will produce analysis tools that practitioners can use (stand-alone, prototype software intended to be integrated into existing system management platforms); as well as guide books for the use of the algorithms; and a project report. In the first year the team will develop the analysis procedures (e.g., congestion “alarms” and monitoring “tools”); and in the second year, the team will fine tune these algorithms and recommend real time strategies that use these tools to address impending or spreading congestion. |