Bayesian neural networks with physics‐aware regularization for probabilistic travel time modeling

Abstract

The integration of data‐driven models such as neural networks for high‐consequence decision making has been largely hindered by their lack of predictive power away from training data and their inability to quantify uncertainties often prevalent in engineering applications. This article presents an ensembling method with function‐space regularization, which allows to integrate prior information about the function of interest, thus improving generalization performance, while enabling quantification of aleatory and epistemic uncertainties. This framework is applied to build a probabilistic ambulance travel time predictor, leveraging historical ambulance data provided by the Fire Department of New York City. Results show that the integration of a non‐Gaussian likelihood and prior information from a road network analysis yields appropriate probabilistic predictions of travel times, which could be further leveraged for emergency medical service (EMS) decision making.

Every Second Counts: A Comprehensive Review of Route Optimization and Priority Control for Urban Emergency Vehicles

“…While these methods can reduce model complexity, they cannot fully capture the vehicle dynamics and features of EMVs and time-dependent features of road traffic flows. To address these limitations, some studies modeled the vehicle dynamics of EMVs and the variance of traffic conditions by proposing comprehensive formulas, including a segmentation function-based model [29], semi-parametric prediction model [30], regression models [31,32], Bureau of Public Road function (BPR function) [33][34][35], and Bayesian neural network [36].…”

“…Relying on overly strong model assumptions [28] and ignorance of the different characteristics of EMVs and SVs will significantly reduce the accuracy of models and algorithms, causing a disconnect between the optimized results and the real-world requirements. In the context of the big data era, future research should fully leverage the advantages of data and concentrate on uncovering the authentic rescue demand characteristics [26] and routing preferences [14,51,52] by collecting and mining actual EMV data (e.g., trajectory data, alarm data from the emergency department) to close the divide between the proposed algorithm and its real-world implementation [27,36].…”