Public Transport Demand Forecasting 2026 — PatSnap Eureka
Public Transport Demand Forecasting: Technology Landscape 2026
From static four-stage travel models to real-time OD matrix prediction powered by deep learning and AFC data — this report maps the patent and literature landscape spanning 2007–2025 for transit operators, urban planners, and MaaS providers.
From Survey-Based Models to Real-Time Data Fusion
Public transport demand forecasting has evolved from static four-stage travel demand models into a dynamic, data-intensive discipline powered by machine learning, real-time sensor fusion, and origin-destination (OD) matrix prediction at fine spatial and temporal granularity. The technology is critical to transit operators, urban planners, and mobility-as-a-service (MaaS) providers seeking to optimize fleet deployment, schedule responsiveness, and infrastructure investment.
The field encompasses four primary technical activities: OD matrix estimation and prediction from historical and real-time data streams; real-time capacity and passenger load forecasting at the vehicle and stop level; network-level demand signal extraction from digital interactions, ticketing systems, and IoT sensors; and agent-based and simulation-driven demand modeling for strategic scenario analysis.
A defining shift is the move from survey-based inputs toward automated fare collection (AFC) data, smart card records, GPS traces, mobile phone signaling, and user app interaction logs as primary demand signals. According to WIPO filing trends, several patents explicitly claim systems that reconstruct OD matrices from electronic ticketing validation data without requiring manual passenger counts. The PatSnap analytics platform indexes this growing body of IP across all major jurisdictions.
The dataset synthesizes patent filings and academic literature spanning 2007–2025. Among the 20 patent records retrieved, China is the most prolific jurisdiction with 11 records, followed by the US with 7, WO with 4, and SG and EP each with 3. No single assignee dominates with more than 3 records in this dataset, indicating that innovation is distributed across a larger number of players than typical semiconductor or pharmaceutical landscapes.
Three Phases of Development: 2007–2025
The dataset reveals three discernible phases of development, from vehicle telemetry and route modeling through machine learning proliferation to the current frontier of deep OD forecasting.
Filing Activity by Innovation Phase
Three phases: Foundational (2007–2016), Development (2017–2021), Advanced/Current (2022–2025) — each defined by distinct technical paradigms.
Key Assignees by Filing Period
Non-Chinese assignees hold multi-jurisdictional portfolios; Chinese academic institutions dominate the 2022–2025 advanced phase.
Four Patent Clusters Defining the Field
The dataset organizes into four distinct technical clusters, each representing a different mechanism for capturing, modeling, and acting on transport demand signals.
OD Matrix Estimation & Deep Learning Demand Prediction
The most active cluster in recent filings. Predicts origin-destination demand matrices at fine spatiotemporal resolution using deep learning architectures. Core mechanisms include multi-period historical OD matrices, graph structures (distance, functional similarity, return-relationship graphs), and attention mechanisms. Beijing Jiaotong University (2023, 2025 CN) applies spatiotemporal multi-graph convolutional neural networks trained on mobile phone signaling data. Shandong University (2023 CN) trains models using multi-cycle historical OD matrices extracting demand-change and urban-area similarity features. Wuhan Institute of Urban Planning (2025 CN) integrates four-stage theory with trip-chain theory for dual-granularity OD matrices. PatSnap analytics tracks this cluster’s rapid growth.
Graph neural networks · Mobile phone signaling · Active CN patentsReal-Time Vehicle Capacity & Stop-Level Demand Prediction
Addresses operational-level forecasting: predicting how many passengers can board at the next stop and anticipating localized demand surges. The mechanism combines real-time vehicle fill-level sensors with expected alight-passenger counts and historical boarding patterns. Gerrit Bohm’s capacity prediction family is the most extensively multi-jurisdictionally protected single technical concept in this dataset, covering predictive boarding capacity display systems across SG, WO, US, CA, AU, and IN. Hitachi’s SG patents (2019, 2021) construct route models combining public transport plan data, movement demand predictions, service congestion, and road traffic congestion to produce multimodal demand-responsive forecasts.
Fill-level sensors · 6-jurisdiction portfolio · Hitachi SGUser Interaction & Digital Signal-Based Demand Inference
Rather than relying on physical sensors or AFC infrastructure, this cluster derives demand signals from digital user interactions — app queries, web searches, social media, and transport information platform logs. Zipabout Limited / Zipabout Local Limited holds EP and WO filings (2019) predicting network demand from user interactions with a transportation information provider system, framing app query logs as demand proxy signals. IBM’s 2016 US patent merges voice-site queries, web queries, and transport route databases to generate supply-demand estimates for multiple transportation aspects in a geographic region. Transit agencies using PatSnap track this emerging signal category.
App query logs · Web search signals · Zipabout EP/WO · IBM USPredictive Fleet Management & Pre-Dispatch Optimization
Bridges demand forecasting and supply-side response: using predicted demand to pre-position vehicles and optimize fleet allocation before requests materialize. Moovit App Global Ltd. (2022 WO, 2023 US) determines spatiotemporal service-need feature vectors, applies time-resolved regression functions to project need vectors for a future period, and uses projections to determine vehicle pre-positioning. Lyft holds three US filings (2020–2022): a geocoded provider reallocation model, a look-ahead/look-behind queue filter pre-dispatch system, and a predictive request model generating customized driver incentive schedules to reduce predicted supply-demand shortfalls. IP strategy teams monitor this cluster for FTO risk.
Pre-dispatch · Queue filters · Lyft US · Moovit WO/USWhere Demand Forecasting Technology Is Being Deployed
From urban fixed-route rail to on-demand microtransit and intercity planning, demand forecasting IP spans a wide range of operational contexts.
What This Landscape Means for IP Strategy and R&D Teams
Five strategic signals drawn from the 2007–2025 patent and literature dataset for transit operators, IP strategists, and technology developers.
ML Model Selection Remains Contested
Random Forest, ANN, DNN, graph convolutional networks, and Holt-Winters time-series methods all appear as claimed approaches across different papers and patents in this dataset. R&D teams should benchmark across architectures on their specific spatiotemporal resolution and data-availability profile before committing to a model family.
AFC Data Is the Most Legally Defensible Proprietary Asset
Multiple patents — from Conduent to South China Agricultural University — build defensible IP around specific methods for reconstructing OD matrices from entry-only or entry-exit smartcard systems. Transit operators that own AFC infrastructure should evaluate whether their data processing pipelines could support protectable innovations.
China-Based Institutions Set the Deep OD Forecasting Frontier
IP strategists entering the public transport demand forecasting space should conduct freedom-to-operate analysis against the growing body of active CN patents from Shandong University, Beijing Jiaotong University, and Wuhan Institute of Urban Planning, particularly for graph-neural-network-based OD prediction architectures.
Five Forward Signals from 2023–2025 Filings
The most recent filings signal where the technical frontier is moving, with Chinese academic institutions leading in architectural sophistication.
Spatiotemporal Multi-Graph Deep Learning for OD Prediction
Beijing Jiaotong University’s 2023/2025 CN patents introduce heterogeneous graph structures — return-relationship, distance, and functional-similarity graphs — combined with attention-based temporal modeling, trained on mobile phone signaling data at intercity scale. This represents the current frontier in architectural sophistication within this dataset. The approach uses random forest for feature selection, constructs multi-graph adjacency matrices, and applies a spatiotemporal multi-graph convolutional neural network. According to IEEE research trends, graph-based spatiotemporal learning is the dominant paradigm in transportation AI.
Heterogeneous graphs · Attention mechanisms · Mobile phone signalingMetropolitan-Circle Integrated Demand Forecasting
Wuhan Institute of Urban Planning’s twin 2025 CN filings combine macro-level four-stage modeling with high-resolution trip-chain theory to simultaneously produce OD matrices at multiple spatial granularities, maintaining temporal continuity across forecast years. This dual-granularity approach addresses a long-standing gap between strategic planning models and operational models. The system integrates motorized and public transport assignment algorithms to derive traffic flow indicators with continuous spatiotemporal consistency.
Four-stage theory · Trip-chain theory · Dual-granularity ODJoint Multi-Modal Demand Prediction
Chongqing University’s 2023/2025 CN patents model taxi and ride-hailing demand as a coupled system, extracting intra-modal and inter-modal spatiotemporal features jointly via adaptive adjacency matrices. This architecture captures inter-modal substitution effects and is applicable to formal public transport systems with competing informal modes. The OECD has highlighted multi-modal integration as a critical gap in urban transport planning.
Adaptive adjacency matrices · Taxi + ride-hailing coupled systemIndividual Trip-Chain OD Prediction & City Management Loops
South China Agricultural University’s 2024 CN patent predicts individual commuter OD data by reconstructing personal trip chains from smartcard history, shifting the unit of analysis from aggregate zones to individual behavior — a prerequisite for personalized MaaS service optimization. Simultaneously, Toyota’s 2024 EP and US filings envision future-demand forecasting as part of a closed-loop city service platform, where demand predictions trigger automated feedback to both service providers (location and timing alerts) and passengers (service availability notifications). The PatSnap Trust Center provides data governance guidance for operators building on individual-level data.
Individual trip chains · Smartcard history · Toyota city platform · Closed-loop feedbackWho Holds the IP in Public Transport Demand Forecasting
| Assignee | Jurisdiction | Filing Period | Technical Focus | Status |
|---|---|---|---|---|
| Gerrit Bohm (individual) | SG, WO, US, CA, AU, IN | 2016–2021 | Vehicle capacity prediction, boarding display systems | Active (multi-jurisdiction) |
| Lyft, Inc. | US | 2020–2022 | Geocoded provider reallocation, queue-filter pre-dispatch, driver incentive scheduling | Active / Pending |
| Hitachi, Ltd. | EP, SG | 2019–2021 | Integrated transit demand prediction devices, multimodal route modeling | Active |
| Beijing Jiaotong University | CN | 2023–2025 | Intercity OD prediction, multi-graph GCN, attention mechanisms, mobile phone signaling | Active |
| Wuhan Institute of Urban Planning | CN | 2025 | Metropolitan-area demand forecasting, four-stage + trip-chain integration | Active |
| Moovit App Global Ltd. | WO, US | 2022–2023 | Spatiotemporal service-need vectors, time-resolved regression pre-positioning | Active |
| Toyota Motor Corporation | US, EP | 2024 | City management support, integrated demand-supply feedback loops | Active |
| Ford Global Technologies, LLC | US, DE | 2022 | Mobility service demand prediction, DRT corridor identification | Active |
Public Transport Demand Forecasting — key questions answered
Modern systems use automated fare collection (AFC) data, smart card records, GPS traces, mobile phone signaling, and user app interaction logs as primary demand signals, replacing traditional survey-based inputs.
Chinese academic institutions dominate active patent generation in deep-learning-based OD demand forecasting, including Beijing Jiaotong University, Shandong University, and Wuhan Institute of Urban Planning, with filings in 2023–2025.
Gerrit Bohm’s capacity prediction for public transport vehicles family is the most extensively multi-jurisdictionally protected single technical concept, with national phase entries across SG, WO, US, CA, AU, and IN.
Lyft holds three US filings (2020–2022) using look-ahead and look-behind queue models comparing projected provider queues to queue capacity to determine pre-dispatch counts, and trains predictive request models to identify supply-demand gaps by geographic area and time period.
Key emerging directions include spatiotemporal multi-graph deep learning for OD prediction, metropolitan-circle integrated demand forecasting combining four-stage and trip-chain theory, joint multi-modal demand prediction for taxis and ride-hailing, individual-level trip-chain OD prediction from smartcard history, and integrated city management demand-supply feedback loops as seen in Toyota’s 2024 EP and US filings.
Multiple patents from Conduent to South China Agricultural University build defensible IP around specific methods for reconstructing OD matrices from entry-only or entry-exit smartcard systems, making AFC infrastructure a legally defensible proprietary data asset.
PatSnap Eureka searches patents and research literature to answer instantly.