How AI Powers Real‑Time Bus and Subway Predictions in Smart Cities

Real‑time Transit Uncertainty

Urban public transportation faces unpredictable delays that affect commuters' schedules and mood. Traditional static timetables struggle to adapt to rapid changes, prompting the need for technology that makes each departure more reliable.

Gaode’s Real‑time Bus and Subway Features

🚌 Real‑time Bus – Tracks vehicle locations, predicts congestion‑induced delays, and eliminates waiting anxiety.

🚇 Real‑time Subway – Shows train operation status, station arrival countdowns, and optimizes transfer efficiency.

These functions rely on a city‑wide solution that integrates data sensing, intelligent computation, and scenario services.

Challenges in Real‑time Bus Prediction

Aleatoric uncertainty in departure times – Actual departures often deviate from schedules due to random factors.

Variable city traffic conditions – Congestion, dedicated bus lanes, and non‑stationary patterns during holidays affect arrival times.

Traffic signal phases – Missed green lights add extra waiting time.

Technical Solution for Bus Prediction

Large‑scale Spatio‑temporal Pre‑training

The model learns long‑range dependencies from historical data, capturing road conditions, traffic‑signal patterns, and passenger flow to support downstream tasks.

Prompt‑based Knowledge Adaptation

Spatio‑temporal Prompt Embedding integrates the pre‑trained encoder into the main network, dynamically focusing on critical information and improving arrival‑time accuracy.

Trajectory Representation Learning

By learning vehicle trajectory patterns, the system better infers operational intent and enhances state recognition.

Challenges in Real‑time Subway Prediction

Data sparsity – Demand varies across regions and times.

Data noise – Random user behavior introduces noise.

Complex temporal patterns – Special services and positioning errors make sequence modeling difficult.

Technical Solution for Subway Prediction

Multimodal Time‑Series Large Model

Similar to visual‑language models, a Time Series Language Model (TSLM) aligns temporal data with semantic space, enabling combined training with large language models for deeper understanding.

Activating Inference Capabilities

Reinforcement learning enhances TSLM’s reasoning, allowing precise detection of subway arrival signals and timely alerts.

User‑Centric Features of Real‑time Subway

In‑station guidance – Optimal transfer paths with walking distance and time estimates.

App‑based boarding code – Seamless “search‑time‑scan” workflow.

Smart alighting reminders – Accurate notifications for getting off, reducing anxiety.

Comfort‑focused upgrades – Clear labeling of cooler or warmer carriages.

By delivering transparent, data‑driven insights, Gaode turns uncertain travel into a simple, confident experience.