Prediction of road traffic flow applying Long Short-Term Memory Model considering impact of COVID-19 in Toyota City

Due to various changes during the COVID-19 pandemic, special changes of road traffic flow are assumed. Changes of detected road traffic flow (DRTF) compared to that of 2019 under the same conditions in Toyota city are analyzed firstly. Generally, the DRTF decrease. Monthly change rate of the DRTF fluctuated during 2020 in 83.6%∼98.3%, however, they keep relatively stable during 2021 in 88.7%∼93.2%. Change rate of one-day-average DRTF for different weekdays, and for three long holidays also have different trends in 2020 and 2021. Moreover, change rate of one-day-average DRTF for different time of state of emergency declarations (SED) have special characteristics. Regarding the analysis above, a Long Short-Term Memory (LSTM) Model which consider impact of COVID-19 is developed to predict one-day DRTF. Sequence-to-sequence (StS) model is introduced, one-to-one and many-to-one models is designed separately to do the prediction. The results demonstrate that MAE, MAPE, and RMSE of one-to-one model is better than many-to-one model, although relationship of DRTF in one week is considered in many-to-one model.