Predicting vacant parking space availability zone-wisely: a densely connected ConvLSTM method

Precise prediction on vacant parking space (VPS) information plays a vital role in intelligent transportation systems for it helps drivers to find parking space quickly to reduce unnecessary waste of time and excessive environmental pollution. By analysing the historical VPS data for an gridded zone, we find that for the number of VPSs, there is not only a solid temporal correlation within the same parking lot, but also an obvious spatial correlation between different parking lots. This paper proposes a hybrid deep learning model, namely the ConvLSTM-DN model, to predict the VPS availability zone-wisely for all parking lots in a target zone. Specifically, taking the gridded two-dimensional historical VPS data in the target zone as the input, the proposed ConvLSTM-DN model can capture the spatial-temporal correlations and make accurate short-term (within 30 minutes) and long-term (over 30 minutes) predictions in the number of VPSs. The ConvLSTM-DN model's performance is evaluated with practical data collected from 9 public parking lots in Santa Monica. The results show that the ConvLSTM-DN model can achieve considerably high accuracy in both short-term and long-term predictions. We also compare the ConvLSTM-DN with an sufficiently fine-tuned Long Short-Term Memory (LSTM) model. The results demonstrate that the ConvLSTM-DN is superior to the latter, especially in long-term predictions.