Long Short-Term Memory Neural Network for Traffic Speed Prediction of Urban Expressways Using Floating Car Data

Recently, accurate and rapid prediction of traffic speed has become a research hotspot in the current applied traffic field. Based on deep learning models, especially Long Short-Term Memory neural network (LSTM), real-time traffic speed prediction of the urban expressway has been a new challenge. Since it takes a long time to extract traffic parameter by map matching method, we present a novel grid model to rapidly derive a series of traffic parameters from the floating car data (FCD). To improve the prediction accuracy, we consider the spatial–temporal characteristics of traffic speed (i.e. upstream and downstream speed, historical average speed, etc.). To verify the validity of the model of the proposed model, we employ 40-day FCD to train and test the model. Our final result, compared with other machine learning methods, the LSTM model has advantages of accuracy and stability, which could facilitate the prediction of the traffic speed and the traffic operation performance.