Limited parking can put a damper on any outing. Travel apps can only help so much when it comes to traffic and popularity of a certain location. Driving around aimlessly doesn’t always yield results, but a new solution is on the horizon.
Researchers at Carnegie Mellon University (CMU) recently published a paper describing an innovative AI system that could predict parking availability in real time. The researchers aimed their focus on monitoring parking meter transactions to accurately predict open parking spaces. It’s estimated that 95 percent of on-street paid parking is regulated by meters.
“In this study, we adopt the data-driven approach by incorporating multiple traffic-related sources, in terms of both real-time and historical data, including parking occupancy, traffic conditions, road characteristics, weather, and network topology,” say CMU researchers. “It ultimately predicts (or forecasts) short-term parking occupancy via a deep neural network method.”
Using a graph convolutional neural network, an algorithm that operates on nodes, edges, properties, and additional graph structures, the team was able to show the statistical relationship between parking locations, traffic flow, parking demand, road links, and parking blocks. In conjunction with a recurrent neural network with long-short-term memory (LSTM), an AI algorithm adept in learning long-term dependencies, and a multi-layer decoder, the system pulls parking information from various data sources including parking meter transactions, traffic speed, weather conditions, as well as output occupancy forecasts.
With a focus area of downtown Pittsburgh, researchers took note of 97 on-street parking meters spread across 36 blocks. The connected car company, Inrix’s Traffic Message Channel, Weather Underground’s API, and Pittsburgh Parking Authority provided traffic speed data and hourly weather reports.
During testing, the model was able to outperform other baseline methods when predicting parking occupancies 30 minutes in advance. The researchers credit the AI’s exceptional performance to weather and traffic speed data, more specifically the weather data that bolstered prediction accuracy in recreational areas.
Co-authors for the paper say, “In general, lower prediction errors are received on blocks with larger parking capacities. It is no surprise as higher parking capacities usually result in lower variances in occupancy rate the model performs better on business districts … [P]arking demand in business districts usually has strong daily patterns, and is more resilient to impacts from unusual scenarios such as hazardous weather and special events, which has made prediction more efficient.”
Going forward, the researchers hope to develop a model that incorporates events, road closures, incidents, and traffic counts into its data inclusion.