A deep learning model is applied for predicting block-level parking occupancy in real time. The model leverages Graph-Convolutional Neural Networks (GCNN) to extract the spatial relations of traffic flow in large-scale networks, and utilizes Recurrent Neural Networks (RNN) with Long-Short Term Memory (LSTM) to capture the temporal features. In addition, the model is capable of taking multiple heterogeneously structured traffic data sources as input, such as parking meter transactions, traffic speed, and weather conditions. The model performance is evaluated through a case study in Pittsburgh downtown area. The proposed model outperforms other baseline methods including multi-layer LSTM and Lasso with an average testing MAPE of 10.6% when predicting block-level parking occupancies 30 minutes in advance. The case study also shows that, in generally, the prediction model works better for business areas than for recreational locations. We found that incorporating traffic speed and weather information can significantly improve the prediction performance. Weather data is particularly useful for improving predicting accuracy in recreational areas.

Over 95% of on-street paid parking stalls are managed by parking meters or kiosks. By analyzing meter transactions data, this paper provides a methodology to estimate on-street timevarying parking occupancy and understand payment behavior in an effective and inexpensive way. We propose a probabilistic payment model to simulate individual payment and parking behavior for each parker. Aggregating the payment/parking of all transactions leads to timevarying occupancy estimation. Two data sets are used to evaluate the methodology, parking spaces near Carnegie Mellon University (CMU) campus, and near the Civic Center in San Francisco. The proposed model generally provides reliable estimations of occupancies at a low error rate and substantially outperforms other naive models in the literature. From the results of the experiments we find that people generally tend to slightly underpay in CMU area, whereas for Civic Center area, payment behavior varies by time of day and day of week. For Fridays, people generally tend to overpay and stay longer in the mornings, compared to underpaying and parking for shorter durations in the late afternoons. Parkers' payment behavior, in general, is more variable and noisier around Civic Center than around CMU. Moreover, we explore the effective granularity, defined as the highest spatial resolution for this model to perform reliably. For CMU areas, the effective granularity is around 10-20 spaces for each block of streets, while it is 150-200 spaces for the Civic Center area due to more random parking behavior.

Travel time on a route varies substantially by time of day and from day to day. It is critical to understand to what extent this variation is correlated with various factors, such as weather, incidents, events or travel demand level in the context of dynamic networks. This helps a better decision making for infrastructure planning and realtime traffic operation. We propose a data-driven approach to understand and predict highway travel time using spatio-temporal features of those factors, all of which are acquired from multiple data sources. The prediction model holistically selects the most related features from a high-dimensional feature space by correlation analysis, principle component analysis and LASSO. We test and compare the performance of several regression models in predicting travel time 30 min in advance via two case studies: (1) a 6-mile highway corridor of I-270N in D.C. region, and (2) a 2.3-mile corridor of I-376E in Pittsburgh region. We found that some bottlenecks scattered in the network can imply congestion on those corridors at least 30 minutes in advance, including those on the alternative route to the corridors of study. In addition, real-time travel time is statistically related to incidents on some specific locations, morning/afternoon travel demand, visibility, precipitation, wind speed/gust and the weather type. All those spatio-temporal information together help improve prediction accuracy, comparing to using only speed data. In both case studies, random forest shows the most promise, reaching a root-mean-squared error of 16.6% and 17.0% respectively in afternoon peak hours for the entire year of 2014.

No abstract