Adaptive Domain Decomposition Physics-Informed Neural Networks for Traffic State Estimation with Sparse Sensor Data

2026-05-08 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Autonomous Vehicles & Smart Transportation · Depth: Expert, medium

Summary

Adaptive Domain Decomposition Physics-Informed Neural Networks (ADD-PINN) is a two-stage, residual-guided framework designed for Lighthill-Whitham-Richards (LWR) model-based offline speed-field reconstruction from sparse fixed sensor data. This method addresses the challenge of traditional PINNs over-smoothing shockwaves. ADD-PINN first trains a coarse global PINN, then uses its spatial residual profile to define subdomain boundaries and initialize child subnetworks. A data-driven shock indicator allows for a single-domain fallback when localized transition evidence is weak. Evaluated on the I-24 MOTION dataset over five days, five sensor configurations, and ten seeds (1,500 runs), ADD-PINN achieved the lowest relative L2 error in 18 of 25 configurations and 14 of 15 sparse-sensing cases. It also trained 2.4 times faster than the extended PINN (XPINN) baseline. Supplementary NGSIM experiments showed the shock indicator suppressing decomposition, with the single-domain fallback performing best, supporting spatial-only decomposition for fixed-sensor traffic reconstruction.

Key takeaway

For traffic engineers and AI scientists working on real-time traffic state estimation with sparse sensor networks, ADD-PINN offers a robust solution for offline speed-field reconstruction. Its ability to accurately capture shockwaves and train faster than existing methods means you can achieve higher fidelity models with reduced computational overhead. Consider integrating this adaptive domain decomposition approach to improve the accuracy and efficiency of your traffic modeling efforts, especially in scenarios with localized transition regions.

Key insights

ADD-PINN improves traffic state estimation from sparse sensor data by adaptively decomposing domains based on residual errors.

Principles

• Residual-guided decomposition enhances PINN accuracy.
• Spatial-only decomposition is effective for fixed-sensor traffic.
• Shock indicators prevent unnecessary domain decomposition.

Method

ADD-PINN trains a global PINN, uses its spatial residual profile to place subdomain boundaries, and initializes child subnetworks. A shock indicator enables single-domain fallback when transition evidence is weak.

In practice

• Apply ADD-PINN for traffic speed-field reconstruction.
• Consider spatial decomposition for sparse fixed sensor data.
• Use residual profiles to guide domain partitioning.

Topics

• Adaptive Domain Decomposition PINN
• Traffic State Estimation
• Sparse Sensor Data
• Lighthill-Whitham-Richards Model
• Physics-Informed Neural Networks

Code references

• thuml/PhySense

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Physics-Informed Neural Network Modeling of Biodegradable Contaminant Transport through GCL/SL Composite Liners — Hard-constrained PINNs significantly improve accuracy and stability for modeling contaminant transport in composite liners.
• Overcoming the Limits of Finite Difference Method; Physics-Informed Neural Network for Noisy High-Dimensional Heat Diffusion — PINNs overcome classical numerical method limitations in noisy, high-dimensional heat diffusion, maintaining accuracy and efficiency.
• Dual-Network PINNs for Optimal Control: A Reproducible Benchmark on the Mass-Spring-Damper System — A dual-network PINN formulation achieves high accuracy in optimal control by exactly enforcing boundary conditions, offering a clear, reproducible alternative to classical methods.
• A Koopman-PINN Framework for Epidemic Models: Parameter Inference and Forecasting — The K–PINN framework combines Koopman linearization with physics-informed learning for robust epidemic modeling and forecasting.
• Transferable Physics-Informed Representations via Closed-Form Head Adaptation — Pi-PINN uses transferable representations and closed-form head adaptation for fast, accurate PDE solutions without extensive training data.
• WaveGraphNet: Physics-Consistent Guided-Wave Damage Localization through Coupled Inverse-Forward Graph Learning — Coupling inverse graph localization with a physics-consistent forward model improves damage localization extrapolation in unseen regions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.