Metropolis-Scale Resilient and Trustworthy Traffic Flow Inference Using Multi-Source Data

Inferring traffic states across a network is crucial for intelligent transportation systems but is complicated by limited data and disturbances in sensor networks. The Task-Aware Attentive Neural Process (TA-ANP) was created as a probabilistic framework that combines floating car data (FCD) and sparse fixed-detector measurements to enhance global traffic state inference (GTSI).

By treating GTSI as a stochastic process, TA-ANP utilizes meta-learning to quickly adapt to changes without needing retraining. It features a multi-query attention module that addresses three GTSI sub-tasks while reducing interference between them.

The framework also incorporates Monte Carlo Dropout for uncertainty quantification, capturing both aleatoric (inherent randomness) and epistemic (lack of knowledge) uncertainties. Testing on the Metropolitan Multi-Source Traffic Dataset (MMTD) demonstrated that TA-ANP achieved state-of-the-art results, allowing for better sensor placement with fewer deployments.

Its resilience to disturbances and adaptability to new configurations make it a valuable tool for urban traffic management.