mobility-aware demand forecasting

This study addresses the limitations of existing mobile cellular traffic load prediction methods, which often underperform due to insufficient modeling of user dynamic behavior. Focusing on highway scenarios, this work proposes a machine learning–based joint prediction model that, for the first time, integrates crowd dynamics—capturing both user count and mobility—as a key input feature alongside historical traffic time-series data. Departing from prior approaches that rely primarily on increasingly complex models, the proposed method underscores the fundamental importance of high-quality, behaviorally informed data for enhancing prediction accuracy. Experimental results demonstrate that incorporating crowd dynamics alone improves prediction accuracy by approximately 60%, substantially outperforming baseline methods.

To address the insufficient accuracy of base station traffic load forecasting—which hinders smart city and intelligent road services—this paper proposes a multi-source time-series forecasting method incorporating prior knowledge from traffic flow dynamics. For the first time, crowd mobility data—including vehicle count and speed—are leveraged as core external priors, replacing conventional approaches that rely solely on endogenous network time series or static points-of-interest (POIs), thereby revealing the underlying human mobility mechanisms driving traffic load generation. Our framework integrates multi-source temporal alignment, sliding-window modeling, adaptive normalization, and error-weighted regression. Evaluated on real-world urban datasets, it achieves up to 56.5% reduction in prediction error, significantly outperforming both pure time-series models and POI-augmented baselines. The source code and visualization results are publicly available.

This study addresses the challenge of insufficient accuracy in urban shared micromobility demand forecasting, which hampers efficient fleet dispatching and traffic management. To this end, we propose a lightweight gradient boosting framework that, for the first time, enables high-precision demand prediction across multiple temporal granularities—from five minutes to one hour—by effectively integrating spatiotemporal and contextual features while supporting edge deployment. Evaluated on real-world datasets of electric scooters and e-bikes from five major cities, our approach significantly outperforms existing methods and generative AI models, demonstrating a strong capability to capture complex mobility dynamics. The proposed framework thus offers an efficient, scalable solution for data-driven decision support in sustainable urban transportation systems.

This work addresses the challenge of modeling short-term fluctuations and uncertainty in high-resolution (15-minute) docked bike-sharing demand forecasting. To this end, we propose T-STAR, a two-stage spatiotemporal adaptive framework that first captures stable hourly demand patterns and then integrates recent demand dynamics with real-time contextual signals—such as metro passenger flows—via a Transformer architecture to produce probabilistic short-term predictions. Our approach innovatively decouples long-term regularities from short-term perturbations, introduces context-aware hierarchical spatiotemporal modeling, and enables zero-shot transfer to unseen regions. Evaluated on the Washington D.C. Capital Bikeshare dataset, T-STAR consistently outperforms state-of-the-art methods across both deterministic and probabilistic metrics, demonstrating strong spatiotemporal generalization and robustness.

This paper addresses the potential and challenges of leveraging large language models (LLMs) for traffic mobility analysis—specifically time-series forecasting. Methodologically, it presents a systematic survey and framework development: (1) it comprehensively maps the research landscape of LLMs in traffic forecasting, clarifying integration paradigms with conventional time-series models, data adaptation bottlenecks, and domain-knowledge injection mechanisms; (2) it proposes the first taxonomy of LLM-based methods for traffic mobility prediction, covering multi-source data encoding, prompt engineering, and model-cooperative modeling. Key limitations identified include inadequate semantic-temporal alignment, limited real-time inference capability, and weak interpretability. The work further outlines scalable technical pathways to overcome these constraints. Collectively, the study establishes a theoretical foundation and practical roadmap for deeply integrating LLMs into intelligent traffic forecasting systems.

This work addresses the limitation of existing mobile traffic forecasting models, which capture only static long-term temporal patterns and fail to model the complex interactions between traffic dynamics and adaptive network parameter adjustments. To overcome this, we propose MobiWM, a world model for mobile networks that treats traffic as a system state and explicitly models its dynamic evolution in response to network actions—such as transmit power, azimuth, and mechanical/electrical tilt—while integrating multimodal environmental context from both images and sequential data. MobiWM enables open-ended, continuous action trajectory rollouts over unlimited time horizons and, for the first time, introduces world models to mobile traffic extrapolation by constructing an explorable counterfactual simulation environment. Evaluated on real-world variable-parameter data spanning 9 regions and 31,900 cells, MobiWM significantly outperforms existing methods in distributional fidelity and effectively supports downstream reinforcement learning optimization, paving the way for digital twin–driven wireless network management.

This work addresses the challenge of accurately modeling dynamic and spatially heterogeneous cellular traffic—particularly in emerging regions with scarce data—where conventional base station deployment and operation strategies fall short. We propose NetSpatial, the first system to integrate spatial conditional generative models into cellular traffic forecasting. By fusing multimodal urban context such as satellite imagery and points of interest, and employing a multi-level flow-matching architecture that decouples periodic patterns from residual dynamics, NetSpatial enables direct long-horizon traffic generation. The framework unifies “what-if” analysis for deployment planning and “what-to-do” optimization for operational decisions. Evaluated on real-world data, it reduces Jensen–Shannon divergence by 29.44%, demonstrates zero-shot generalization across cities, and achieves 16.8% energy savings through base station sleep scheduling and load balancing while maintaining quality of experience for over 80% of users.

This study addresses the growing demand for mobility insights across urban planning, transportation, and retail by proposing an end-to-end urban mobility analytics framework. Built upon a reusable modular architecture, the framework integrates high spatiotemporal resolution mobility modeling with multi-scenario business applications. It establishes a closed-loop pipeline—from raw geolocation data to strategic insights—through anonymization, ETL workflows, BigQuery-based data management, Vertex AI–driven model training, and Power BI visualization. The system effectively supports diverse analytical tasks, including traveler profiling, trajectory mining, catchment area analysis, traffic anomaly detection, and origin–destination pattern recognition, thereby delivering scalable and efficient decision support for both urban governance and commercial strategy.

This study addresses the limitations of existing electric vehicle (EV) charging demand forecasting, which often relies on outdated data and fails to capture the scale and behavioral heterogeneity of modern charging networks. The authors construct a large-scale longitudinal dataset of EV charging activity in Scotland spanning 2022–2025 and introduce the first unified probabilistic framework that models charging demand as a spatiotemporal latent Gaussian field, integrating spatial dependencies, temporal dynamics, and covariate effects. Leveraging integrated nested Laplace approximation (INLA) for efficient Bayesian inference, the proposed approach achieves site-level prediction accuracy comparable to state-of-the-art machine learning models while providing reliable uncertainty quantification and interpretable decomposition of spatiotemporal components. This enables risk-aware decision support for power grid dispatch and infrastructure planning. The publicly released dataset establishes a new benchmark for the field.