The 20th and 21st joint COTA International Conference of Transportation Professionals (CICTP 2020-21)

The Imbalance between travel supply and travel demand will result in traffic congestion in urban street networks, and predicting traveling speed well has the potential to mitigate congestion since the predicted speed helps travelers make optimal paths to avoid congested streets. The paper proposes a multilayer long short-term memory (LSTM) model and a multilayer gated recurrent unit (GRU) model which are time sequence deep learning models for predicting street speed since recurrent neural network (RNN) models can solve time sequence problems. Considering the impact of external factors like weather condition, the models take the factors into account as variables. An urban street in Manhattan is taken as the case to research the efficiency of the multilayer LSTM model and the GRU model. The result suggests that both deep LSTM model and deep GRU model outperform other conventional models according to the error measurements. To research further, the deep learning model will be studied whether it can predict the speed of the all the streets in the network in the future.

CICTP