New Generation Smart Parking Platform: Data-driven Parking Lot Assignment Approach Considering Drivers’ Heterogeneous Preferences

Efficient parking lot assignment is a persistent challenge for smart parking platforms that manage a set of parking lots within an urban area. Traditional parking management platforms focus on maximize the system revenue or utilization of parking resources. However, these platforms often overlook drivers’ personalized tolerance for walking distances after parking in the assigned parking lot, resulting in low demand fulfillment and drivers’ satisfaction. This paper proposes a parking lot assignment approach for new generation of smart parking platform, which incorporates an online estimation of each driver’s acceptance probability of walking distance. A discretized learning structure with interpolation and monotonic update rules is employed to infer the acceptance probability from online assignment feedback. A novel measure for alternative parking lots is introduced and integrated into the objective function to guide more informative assignments. The resulting assignment problem is formulated as an optimization assignment model and solved iteratively at each time step. Numerical experiments conducted in a synthetic urban grid demonstrate that the proposed method (C-L) significantly outperforms a distance-minimization baseline, achieving over $90 \%$ of the fulfillment attained by an ideal oracle benchmark. Furthermore, C-L exhibits rapid regret convergence and strong scalability across varying demand levels. These findings highlight the importance of incorporating behavioral preferences and adaptive learning in parking lot assignment for region-lot integrated network platforms.