Developing risk-informed speed limits against single-vehicle crashes by exploiting an augmented reliability problem with multi-fidelity enhancement

Excessive speed has been blamed as a primary contributory and aggravating factor of single-vehicle crashes (SVCs), especially for vehicles under adverse driving environments (e.g., slippery road surface and strong wind). Rational advisory speed limits (SLs) that can adapt to time-varying environments become critical to ameliorating SVC risks while maintaining optimal traffic mobility. Existing SVC modeling as well as associated SL strategy are either over-simplified with questionable accuracy or too complicated and computationally expensive to accommodate timely risk prediction and potential mitigation. In this regard, a novel risk-informed SL strategy against SVCs is developed. Rather than performing reliability-based optimization for SL with traditionally repeated reliability analyses, an augmented reliability problem (ARP) is formulated. The accuracy is guaranteed by exploiting the ARP through the efficient non-parametric stochastic subset optimization with a high-fidelity SVC model, and a low-fidelity SVC model is incorporated to further improve efficiency. Demonstrations are conducted based on several examples designed with AASHTO Green Book. The results indicate that, in degraded driving environments, the original SL can induce increasing SVC risks, and the optimal SL with acceptable SVC risks decreases significantly. The proposed method can facilitate a reliable SL modulation that can quickly adapt to the changing driving environment with only a small number of high-fidelity simulations. It bears great potential to build an intelligent and proactive traffic management system against SVCs with well-informed and consistent risk levels in response to forthcoming hazardous weather events.