Real-time redesign of public transport lines under fluctuating traffic conditions

Public Transport (PT) lines are traditionally designed to optimize performance under nominal traffic conditions. In practice, however, operating conditions frequently deviate from the nominal ones, leading to substantial performance deterioration. Existing adaptation mechanisms typically rely on corrective and reactive interventions, such as stop-skipping, which prove insufficient under large or recurrent traffic fluctuations. In such contexts, incremental adjustments are inadequate, and a structural redesign of the network becomes necessary. This paper introduces a real-time optimization method for proactively redesigning PT networks to maintain high levels of user- and operator-oriented performance under pronounced traffic fluctuations. We adopt a predict-then-optimize paradigm in which PT lines are reconfigured based on predicted traffic conditions via an algorithm based on the Non-dominated Sorting Genetic Algorithm III (NSGA-III). To ensure operational implementability, we explicitly constrain structural deviations by enforcing high Jaccard similarity between the baseline and redesigned networks, thereby preserving network continuityand limiting operational disruption. To account for the effect of prediction inaccuracies on redesign quality, we construct a statistical model of the errors made by a well-established deep learning-based prediction model (Diffusion Convolutional Recurrent Neural Network) trained on real-world data. Computational results on Mandl’s benchmark network, under high traffic fluctuations, show that at least 77.9% of OD pairs experience travel time reductions of at least 20%, alongside operational cost savings of up to 37%, while retaining 79% of the original network structure. Experiments on the large-scale Beijing network further demonstrate average travel time improvements of up to 20.5% and operational cost reductions of 15.6%, while maintaining more than 60% route overlap with the incumbent configuration. These findings demonstrate that proactive, prediction-driven redesign can substantially improve performance, compared to current inflexible PT designs, thereby motivating a transition from current static planning paradigms toward continuous and adaptive PT network design.