Improving Parking Availability Information Using Deep Learning Techniques

Abstract

Urban traffic currently affects the quality of life in cities and metropolitan areas as the problem becomes ever more aggravated by parking issues: congestion increases due to individuals looking for slots to park their vehicles. An Internet of Things approach allows drivers to know the state of the parking system in real time through wireless networks of sensor devices. This work focuses on studying the data generated by parking systems in order to develop predictive models that generate forecasted information. This can be useful in improving the management of parking areas, especially on-street parking, while having an important effect on urban traffic. This research begins by looking at the state of the art in predictive methods based on machine learning for time series. Similar studies and proposed solutions for parking prediction are described in terms of the technology and current state-of-the-art predictive models. This paper then introduces the recurrent neural network methods that were used in this research, namely Long Short-Term Memory and Gated Recurrent Unit, as well as the models developed according to real scenarios in different cities. In order to improve the quality of the models, exogenous variables like hourly weather and calendar effects are taken into account, and the baseline models are compared to the models that used this information. Finally, the preliminary encouraging results are described, followed by suggestions for corresponding future work.