Parking Cluster Zones

• Parking Cluster Zones are spatial groupings of parking facilities with shared demand and management patterns, defined through data-driven methods.
• PCZ frameworks utilize advanced algorithms such as k-means, Gaussian mixture models, and optimization techniques to improve pricing strategies and prediction accuracy.
• Integrative approaches combining real-time sensing, machine learning, and policy frameworks enable dynamic pricing, congestion mitigation, and efficient urban parking management.

Parking Cluster Zones (PCZs) are defined as spatial groupings of parking facilities—ranging from curbside segments to off-street lots—that share common usage patterns, demand profiles, or operational management. The core rationale underlying PCZs is that treating parking not as isolated spots but as spatially and functionally correlated clusters can enable more sophisticated demand modeling, pricing, optimization, and management strategies. PCZs serve as the administrative and analytical entities for several advanced frameworks in dynamic parking pricing, spatio-temporal demand modeling, availability prediction, congestion mitigation, and smart allocation in urban settings.

1. Motivation and Conceptual Definition

Parking Cluster Zones emerge from documented empirical heterogeneity in spatio-temporal parking demand, network effects, and operational challenges in urban mobility. Rather than impose arbitrary policy zones or treat each parking facility individually, PCZs are constructed via data-driven methodologies—typically clustering algorithms or optimization models—that seek zones with high spatial and temporal correlation in parking occupancy, demand, or contextual features such as proximity to transit, land use, or travel demand characteristics (Fiez et al., 2017, Huang et al., 4 Sep 2025, Huang et al., 2 May 2024).

Key attributes defining a PCZ include:

• Intra-zone demand homogeneity: facilities within a PCZ exhibit similar occupancy and usage patterns, increasing the effectiveness of joint management.
• Spatial contiguity: clusters are contiguous or near-contiguous spatially, often shaped by city block topology or functional urban boundaries.
• Managerial tractability: clusters support coordinated control actions (pricing, assignments, policy interventions) at a scale compatible with urban governance.

2. Mathematical and Computational Formulations

The definition and operationalization of PCZs are instantiated through clustering and mixed-integer optimization methods. Typical approaches include:

• K-means clustering on occupancy data, spatial features, or traffic inflow/outflow patterns, with city-aware metrics such as Minkowski distance to capture non-Euclidean urban geometry. Once clusters are formed, each becomes a PCZ, with boundaries encompassing parking assets with correlated behavior (Huang et al., 4 Sep 2025, Huang et al., 2 May 2024).
• Gaussian mixture models (GMMs) leveraging EM algorithm for unsupervised clustering based on multidimensional feature vectors incorporating location and occupancy. The number of clusters (zones) is selected by model selection criteria such as BIC (Fiez et al., 2017).
• Optimization-based tessellation: Binary integer programming partitions facilities into nonoverlapping, visually disjoint zones, often optimizing a target such as profit, utilization, or operational cost. Discrete variables encode facility-to-zone assignments, with additional variables for pricing or policy settings (Deng et al., 2023).
• Time-unrolling dynamic programming: For dynamic zoning, the assignment of parking assets to clusters may itself evolve over time, with temporal regularization penalizing frequent reconfiguration (Nazir et al., 2022).

These computational frameworks allow both static definition and dynamic adaptation of PCZs as operational needs shift or demand fluctuates.

3. Demand Modeling, Prediction, and Spatial Correlation

A foundational reason for establishing PCZs lies in exploiting spatial and temporal correlations to enhance demand modeling and prediction. Empirical analysis demonstrates:

• Spatio-temporal autocorrelation: Moran’s I and similar metrics reveal strong spatial correlation in occupancy within well-chosen PCZs, in contrast to arbitrary policy zones (Fiez et al., 2017). Temporal consistency metrics, such as cluster assignment repeatability, further establish that cluster-based zoning remains robust over time.
• Multi-source data fusion frameworks: By gathering and integrating parking, transit, taxi, ride-hailing, and traffic data at the PCZ granularity, predictive models can account for complex inter-modal competition, choke points, and event-driven surges. Cluster-wise data fusion enhances feature richness and supports more accurate deep learning models (Huang et al., 2 May 2024, Huang et al., 4 Sep 2025).
• Self-supervised spatio-temporal representation learning: Masking and reconstruction tasks on continuous temporal segments and spatial nodes (within the PCZ) facilitate the learning of spatial structure and temporal dependencies. Excluding neighboring lots’ data within a PCZ robustly degrades predictive performance, quantitatively confirming the need for spatial modeling at the PCZ level (Huang et al., 4 Sep 2025).
• Transformer models: Mechanisms such as dual-branch attention (series/channel) in SST-iTransformer or classical Transformer self-attention facilitate long-term dependency extraction across both time and multiple correlated lots within a PCZ (Huang et al., 4 Sep 2025, Huang et al., 2 May 2024).

The net effect is the significant reduction in MSE, MAE, and MAPE for parking availability predictions when using PCZ-aware data-driven models compared to lot-level forecasting.

4. Dynamic Pricing, Assignment, and Management

PCZs are a natural structure for deploying dynamic pricing and coordinated assignment policies:

• Bi-level Stackelberg dynamic pricing: PCZs enable a parking authority to set real-time prices at the cluster level, guiding demand to achieve occupancy targets (e.g., 85%) while balancing social surplus and revenue (Mackowski et al., 2015). The aggregated occupancy within the cluster is steered by periodically updated prices, with users solving assignment problems based on the composite disutility incorporating walking and driving costs.
• Networked queueing models: Networks of PCZs can be modeled as interconnected queues with stochastic rejection flows; price can be controlled locally at each PCZ to maximize utilization while maintaining cruise-generated congestion below defined thresholds. Convex optimization with price elasticities over PCZs enables fine-tuned spatial control (Dowling et al., 2017).
• Assignment and search minimization: Matching models (Hungarian algorithm) applied at the PCZ scale, using effective travel times penalized by historical cluster-level success ratios, dramatically reduce search time and congestion compared to uncoordinated or naïve strategies (Hemmatpour et al., 27 Aug 2025).
• Distributed allocation and reserve management: PCZs facilitate fair, privacy-preserving distributed allocation of premium parking (e.g., through stochastic iterative protocols) and support buffer sizing for uncertainties in shared-use contexts (Griggs et al., 2015).

These models can be tuned for real-time management, policy constraints (e.g., commercial vehicle restrictions), or user preference profiles and demonstrate robust empirical gains in utilization, revenue, and congestion reduction.

5. Sensing, Detection, and Data Infrastructure

PCZ use enhances the feasibility and scalability of sensing and detection strategies:

• Crowdsourcing and mobile sensing: Cohesive clusters benefit from mobile detection approaches, where aggregated information from traversing vehicles is sufficient to support cluster-level control with fewer sensors compared to fixed, dense deployment at each space (Liao et al., 2016).
• Vision-based detection: Computer vision and deep learning, including instance segmentation (Mask R-CNN, Cascade Mask R-CNN, DETR), can detect parking spaces automatically at the cluster level, facilitating PCZ definition and monitoring without intensive human labeling (Almeida et al., 2023, Grbić et al., 2023, Nguyen et al., 7 Jul 2024).
• Self-adjusting bounding boxes and masking: Algorithms that auto-correct for skewed orientations and camera placements can maintain high recall and precision for occupancy at the PCZ scale in real time (Nguyen et al., 7 Jul 2024).
• Data fusion: Integrating multi-modal demand signals (transit, ride-hailing) at the PCZ boundary enables better temporal staleness mitigation, robust measurement, and more actionable predictive outputs for users and agencies (Huang et al., 2 May 2024, Huang et al., 4 Sep 2025).

The PCZ framework streamlines both the operational burden of data acquisition and the system-level fidelity of real-time monitoring.

6. Urban Planning, Policy, and Future Directions

PCZs provide a rigorous, data-driven basis for several aspects of urban mobility and governance:

• Dynamic and adaptive zoning: PCZs facilitate dynamic rezoning and adaptive assignment of curbside uses (e.g., paid parking, loading, transit), maximizing curb valuation under evolving demand and regulatory constraints (Nazir et al., 2022).
• Integration with SAV deployment: In shared autonomous vehicle contexts, PCZs structure the optimal density, location, and relocation of SAV stations and parking clusters, accounting for inter- and intra-zone trip balances and minimizing operating costs (Choi et al., 2022, Kumakoshi et al., 2021).
• Traffic management interface: PCZs can serve as buffer zones for network-level congestion mitigation in CAV-integrated urban traffic, enabling temporary vehicle holding to balance throughput and delay (Liu et al., 5 Dec 2024).
• Urban land use and redevelopment: The freeing up or concentration of parking made possible by PCZ efficiency gains supports land repurposing strategies, enhanced business vitality, and more sustainable spatial allocations (Kumakoshi et al., 2021).
• Policy and revenue optimization: Joint zonification and pricing models demonstrate empirically a significant uplift in system profits and user experience compared to legacy administrative partitions such as zip code boundaries (Deng et al., 2023).

The PCZ concept is expected to play a central role as urban environments move toward data-rich, multi-modal, and highly dynamic mobility paradigms, integrating advanced prediction, optimization, and adaptive management techniques.

---

The Parking Cluster Zone framework thus encapsulates a convergence of spatial-statistical analysis, deep learning, operational research, and sensing for scalable, robust, and efficient urban parking management. Its applications span from fine-grained availability prediction to city-level optimization of parking resources, demonstrating both immediate operational benefits and long-term policy impact.