Predicting vacant parking space availability zone-wisely: a hybrid deep learning approach

Abstract Precise prediction on vacant parking space (VPS) information plays a vital role in intelligent transportation systems for it helps drivers to find the quickly reduce unnecessary waste of time and excessive environmental pollution. By analyzing historical zone-wise VPS data, we that number VPSs, there is not only solid temporal correlation within each lot, but also an obvious spatial among different lots. Given this, this paper proposes hybrid deep learning framework, known as dConvLSTM-DCN (dual Convolutional Long Short-Term Memory with Dense Network), make short-term (within 30 min) long-term (over predictions availability zone-wisely. Specifically, correlations scales, namely 5-min daily-wise lots can be effectively captured by two parallel ConvLSTM components, meanwhile, dense convolutional network leveraged further improve propagation reuse features process. Besides, two-layer linear used extract meta-info promote accuracy. For predictions, methods, direct iterative are developed. The performance model extensively evaluated practical data collected from nine public Santa Monica. results show framework achieve considerably high accuracy both predictions.