Study On The Data-characteristic-driven Short-term Passenger Flow Prediction Of Urban Rail Transit

In recent years,urban rail transit in China has maintained a rapid development momentum due to its unique advantages such as large traffic volume,fast operation speed,safe punctuality,environmental protection and comfort,and effective improvement of urban ground traffic pressure.Urban rail transit passenger flow forecasting has always been one of the key tasks in the entire life cycle of urban rail construction and operation.The prediction results are of great significance to passenger travel,operator decision-making,and the safe operation of urban rail transit.This paper aims to build a high-precision short-term passenger flow prediction model for urban rail transit.First,based on the AFC data of Chengdu rail transit,the pre-processing process of the metro passenger flow data set is introduced in detail.Then,an integrated urban rail transit passenger flow data characteristic analysis framework is proposed,which mainly consists of two parts: essential feature identification and mode feature quantification,in order to explore the stability / non-stationarity,linear / non-linearity,and complexity of urban rail transit short-term passenger flow data.Nature,periodicity,mutation and randomness.Taking Chengdu Metro Station Tianhe Road Station and Tianfu Square Station as an example,the results show that the short-term passenger flow data of urban rail transit is non-stationary,non-linear,highly complex,periodic,abrupt,and random.The analysis of the characteristics of short-term traffic flow data lays the foundation for the selection of passenger flow prediction models.Considering the complex characteristics of short-term passenger flow data of urban rail transit,this paper establishes a short-term passenger flow prediction model based on LSTM / GRU for urban rail transit,and uses Bayesian optimization to adjust hyperparameters.Taking Chengdu Metro Station Tianhe Road Station and Tianfu Square Station as examples,the results show that the model proposed in the paper can better obtain the time series characteristics of short-term passenger flow data of urban rail transit,and the model prediction results can better fit the real data.Finally,in order to verify the superiority of the proposed model,the classic statistical model ARIMA,machine learning model SVR,and traditional RNN model are selected as the benchmark model.The Chengdu rail transit data is also used as an example.The results show the MAE and RMSE of LSTM / GRU.The indicators are smaller than the above benchmark models.