Abstract Manager

Like many cities, Toronto suffers from limited parking availability and high demand, which make searching for parking spaces cumbersome. Rather than spending considerable amount of time in an almost fully occupied parking spot near the desired destination, a less expensive but farther parking spot within walking distance may be an alternative for drivers. 

To balance the demand and the supply of parking spots, this study proposes an adaptive pricing algorithm for on-street parking spots in close vicinity, i.e., increasing the parking cost in a crowded spot and lowering it in a less-congested one. An iterative optimization algorithm is introduced to determine the optimal price for individual parking spots. The number of available parking spots (NAPS) is estimated using the transaction data provided by the Toronto Parking Authority (TPA), which is financially efficient and easily accessible. Exploiting other techniques for estimating NAPS, e.g., installing cameras or sensors, is comparatively expensive and intrusive. The TPA data comprises arrival and departure times, dates, and paid amounts for approximately 1000 parking locations in Toronto during May, June, and July 2019 and 2022.

A crucial part of this research is the development of a Graph Neural Network (GNN). The GNN, whose graph nodes represent parking locations, is specifically trained by the TPA data to predict the NAPS after updating the price profile. Input features for each parking spot are categorized into sociodemographic data from the City of Toronto and amenity data obtained from the Open Street Map. Initial results suggest that the GNN framework captures spatial relationships among parking locations. In the training phase, after 250 epochs, the Root Mean Square Error (RMSE) in estimating the NAPS diminishes from 14.3905  to 5.3243 over the validation set. In the adaptive pricing algorithm, following 400 iterations, the number of neighbourhoods with an unbalanced NAPSs and price of parking spots is decreased from 678 to 364, with price adjustments applied to 235 parking spots.