Gru_Attention_TrafficFlowPrediction

Several methods including GRU Dual Stage Attention Netwrok, GRU, LSTM, TCN were used in this work to predict traffic flow, share and speed on PeMS04 data.

Directory and File Description

Directory:

• data:Raw data
  
• processed: Stored data which have been processed. The process includes data standardization through Z-score and historical features extraction.
  
• *_model: A various of models to predict traffic task that have been trained and persisted in the way of pkl.
  
• re: It stored prediction result in the way of csv, which given by these models.
  

File:

• main: The entry pint of the program
• *_network: The structure of various models
• my_parameter: Some of settings about our models, such as various path definition, history window which decided how long period historical data I used

Prediction Result Display

Model	MAE	R Square
LSTM	27.928	0.875
GRU	25.579	0.911
GRU_Attention	22.626	0.934
GRU_Dual_Stage_Attention	21.913	0.937
TCN_Attention	26.530	0.909

Reference

• [1] An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling
• [2] A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction