G-RGAN: A Spatiotemporal Graph Generative Adversarial Model of Parking Data Recovery
- Resource Type
- Conference
- Authors
- Wu, Weiwei; Peng, Lei
- Source
- 2021 4th International Conference on Intelligent Autonomous Systems (ICoIAS) ICOIAS Intelligent Autonomous Systems (ICoIAS), 2021 4th International Conference on. :181-186 May, 2021
- Subject
- Computing and Processing
Recurrent neural networks
Roads
Urban areas
Buildings
Predictive models
Generative adversarial networks
Data models
time-series data generation
graphic generative adversarial network
spatial dependence
- Language
Recurrent neural networks (RNN) based methods are usually employed to predict or generate time-series data such as traffic flow or parking data. However, the facts show the data of traffic or parking is also impacted by geospatial relationships such as road or parking lot networks. Spatiotemporal graph neural networks are recently brought into focus in the field of traffic flow prediction due to the ability to represent geospatial and temporal characteristics of traffic flow simultaneously. In this paper, the spatiotemporal graph neural networks named G-RGAN are developed for parking data recovery, which is the generative model corresponding to the discriminative model referring to parking prediction. The G-RGAN and other related methods are used to recover parking data of several real parking lots to compare their performance. And the experiment results show that G-RGAN can learn latent characteristics of geospatial influence on parking, thus the quality of generated data is better than the state-of-art baselines.