A Physics-Based Diagnostic Layer for Urban Traffic Network Analysis

Read full text Summarise with VIRAT
Authors: Janos Gabor Melegh
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

A Physics-Based Diagnostic Layer for Urban Traffic Network Analysis From Early Warning System Hypothesis to Validated Network Stress Diagnostics This paper presents a physics-based diagnostic framework for urban traffic network analysis. The project originated with an ambitious goal: developing an Early Warning System (EWS) capable of predicting traffic network collapse through critical slowing down signatures. Through rigorous validation including null tests, no-event baselines, and delayed-event experiments on a calibrated Budapest road network (42 nodes, 59 edges), we demonstrate that the proposed EWS indicators were artifacts of initialization transients rather than genuine precursor signals. However, the validation process yielded a significant positive outcome: a robust diagnostic layer achieving edge stress ratios of 0.69 on real urban networks. The resulting system provides real-time network stress visualization without predictive claims - a traffic X-ray machine that reveals hidden congestion patterns in urban road networks. Keywords: urban traffic networks, network diagnostics, early warning systems, critical slowing down, physics-based modeling, Budapest road network 1. Introduction 1.1 The Reactive Paradigm Problem Current state-of-the-art urban traffic management systems, and recent reinforcement learning approaches (march/2026)—share a fundamental limitation: they are inherently reactive. These systems detect congestion after it forms and respond with signal timing adjustments. None can predict network-level collapse before it occurs. The theoretical foundation for prediction exists. Complex systems theory, particularly Critical Slowing Down (CSD) theory, suggests that systems approaching phase transitions exhibit characteristic early warning signatures: increasing variance, rising autocorrelation, and growing spatial heterogeneity. If urban traffic networks behave as critical systems, these signatures could provide advance warning of cascading congestion. 1.2 Project Objectives System was designed to test this hypothesis rigorously. The project had two parallel objectives: Primary: Develop and validate physics-based Early Warning System (EWS) indicators for traffic network collapse prediction Secondary: Create a robust diagnostic overlay for real-time network stress visualization This paper documents the complete validation journey, including the negative result on EWS and the positive outcome of a validated diagnostic layer. 2. Related Work 2.1 Adaptive Traffic Control Systems SCOOT (Split Cycle Offset Optimisation Technique, 1981) and SCATS (Sydney Coordinated Adaptive Traffic System, 1982) represent the dominant paradigm in deployed traffic management. Both systems use loop detector data to optimize signal timing parameters (split, cycle, offset) with update cycles of several seconds. While reliable and well-documented, they lack any network-level health metrics or predictive components. Surtrac (Carnegie Mellon, Pittsburgh) advances this paradigm with decentralized, forward-looking scheduling that achieves 25-40% wait time reductions. However, its optimization horizon remains corridor-level, without global network health assessment. Recent reinforcement learning approaches (e.g., DeepMind Singapore 2023) demonstrate promising simulation results but operate as black boxes without network-level criticality estimation. 2.2 Critical Slowing Down in Complex Systems CSD theory predicts that systems approaching critical transitions exhibit universal early warning signatures. Scheffer et al. (2009) demonstrated these signatures in ecological systems; Dakos et al. (2012) formalized detection methods. Key indicators include: Increasing variance in system state variables Rising lag-1 autocorrelation Growing spatial heterogeneity Critical slowing of recovery from perturbations Application to traffic networks has been limited to theoretical proposals. This project represents, to our knowledge, the first rigorous empirical validation attempt. 3. Methodology 3.1 System Architecture System implements a layered architecture: Topology Layer: Graph representation (nodes = intersections, edges = road segments) Flow Layer: CTM-inspired density-based dynamics with GRID-style flow balance Metrics Layer: Edge stress ratios, queue accumulation, heterogeneity measures Coarse-Graining Layer: Community detection for zone-level aggregation EWS Layer: Candidate early warning indicators (subsequently invalidated) 3.2 Core Dynamics The simulation updates edge densities (ρ) and node queues (q) at each timestep: ρ(t+1) = ρ(t) + dt × [inflow - outflow + damping × (ρ_eq - ρ(t))] q(t+1) = q(t) + dt × [demand - service_rate] - queue_decay × q(t) 3.2.1 Key Parameters Parameter Symbol Default Value Description Timestep dt 1.0 Simulation timestep Total time T_total 90-120 Simulation duration (steps) Warmup warmup_steps 30 Pre-event equilibration period Control gain control_gain 1.05 Local stabilization strength Damping damping 0.08 Density relaxation rate Queue decay queue_decay 0.05 Queue dissipation rate Overload lambda λ 2.0 Overload sensitivity 3.3 Diagnostic Metrics 3.3.1 Edge Stress Ratio (Primary Diagnostic) edge_ratio = (edges with ρ > threshold) / total_edges Calibrated threshold: 0.25 × capacity (after systematic sweep) 3.3.2 Queue Heterogeneity queue_heterogeneity = std(queue_values) / mean(queue_values) Measures spatial variance in queue accumulation across nodes. 3.3.3 Queue Peak queue_peak = max(mean(queue_values) over time) Maximum network-wide average queue level. 3.4 EWS Candidate Indicators (Tested and Rejected) 3.4.1 D_hist Composite D_hist = var(queue) + var(overload) + mean(|K|) Where K = curvature-like field derived from density gradients. 3.4.2 D_hist_lag1 (Primary EWS Candidate) Lag-1 autocorrelation of D_hist timeseries. CSD theory predicts this approaches 1.0 before critical transitions. 3.4.3 C_hist_roll_var Rolling variance of C_hist = 1/(1 + var(ρ_C)) Where ρ_C = exp(-λ × overload). 3.4.4 d(queue_heterogeneity)/dt Time derivative of queue heterogeneity. Tested as leading indicator of queue peak. 4. Experimental Setup 4.1 Network Topologies 4.1.1 Budapest Pilot Network Real urban network extracted via OSMnx from Budapest, Hungary: Property Value Nodes 42 Edges 59 Network type Core urban subgraph Source OpenStreetMap via OSMnx Simplification simplify=True (standard) 4.1.2 Synthetic 4×4 Benchmark Control network for artifact detection: Property Value Nodes 16 Edges 24-26 Topology Regular grid Purpose Warmup artifact detection 4.2 Event Profiles Four standardized perturbation scenarios: Profile Description Mechanism accident_only Single link capacity reduction 50% capacity on critical edge closure_only Complete link closure 0% capacity on selected edge combined Accident + demand spike Capacity reduction + 30% demand increase demand_spike_only Pure demand increase 30% network-wide demand surge 4.3 Calibration Protocol Systematic grid search over: capacity_scale: [1.0, 0.8, 0.6, 0.55, 0.5, 0.45] demand_scale: [0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5] edge_hot_threshold_ratio: [0.85, 0.7, 0.5, 0.35, 0.25] Selection criterion: Maximize edge_ratio_peak while maintaining stable dynamics. 4.3.1 Final Calibration (Budapest) Parameter Calibrated Value capacity_scale 0.45 demand_scale 2.5 (pilot) / 3.5 (stress test) edge_hot_threshold_ratio 0.25 edge_ratio_peak achieved 0.69 (pilot) / 0.92 (stress) 4.4 Validation Protocol 4.4.1 Null Test Design Critical validation for EWS claims requires demonstrating event-specificity: Standard run: Event at t = warmup + event_base_start (t=40) No-event baseline: Identical parameters, no event injection Delayed-event run: Event at t = warmup + event_base_start + delay (t=60) Pass criterion: If indicator is event-specific, its crossing index must: Differ between standard and no-event runs Shift proportionally with event delay 4.4.2 Robustness Sweep 36 configurations tested: Seeds: [11, 21, 42, 84] Warmup lengths: [20, 30, 40] Community resolutions: [1.0, 1.5, 2.0] All 4 event profiles 5. Results 5.1 Calibration Results 5.1.1 Budapest Pilot Calibration Profile queue_peak queue_area edge_ratio_peak lead_time_D_lag1 accident_only 26.42 24740.7 0.69 +9 closure_only 26.41 24034.2 0.69 +9 combined 26.45 24892.1 0.69 +9 demand_spike_only 26.43 24156.8 0.69 +9 5.1.2 Stress Test Calibration Profile queue_peak queue_area edge_ratio_peak lead_time_D_lag1 accident_only 18.37 142367.7 0.92 +9 closure_only 18.35 141892.3 0.92 +9 combined 18.38 142501.2 0.92 +9 demand_spike_only 18.36 142089.5 0.92 +9 5.2 EWS Validation Results (Negative) 5.2.1 Null Test Results Critical finding: D_hist_lag1 crossing index = 30 in ALL conditions: Condition D_hist_lag1 cross_idx Warmup length Difference Standard (event at t=40) 30 30 0 No-event baseline 30 30 0 Delayed event (t=60) 30 30 0 Interpretation: The crossing occurs exactly at the warmup boundary regardless of event presence or timing. This is conclusive evidence that the signal is an initialization transient artifact, not an event-specific early warning. 5.2.2 Delay Shift Analysis delay_shift = delayed_cross_idx - standard_cross_idx = 30 - 30 = 0 Expected for valid EWS: delay_shift ≈ 20 (matching the event delay) Observed: delay_shift = 0 across all profiles 5.2.3 Event Impact Analysis Queue peak comparison (no-event vs standard): Metric No-event baseline Standard (with event) Difference % Change queue_peak 27.031 27.039 0.008 0.03% Critical implication: Events have negligible impact on network state. If events don't significantly perturb the system, there is nothing to predict. 5.2.4 Artifact Diagnosis Indicator Artifact Type Evidence C_hist_r

Fields of Study & Subjects

Traffic control and management Traffic Prediction and Management Techniques Ecosystem dynamics and resilience

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
30 March 2026
Language
EN
Document type
Preprint