Travel Time Prediction For Subway Passenger Based On Machine Learning

The travel time of subway passengers is one of the important indexes to measure the service level of subway.Accurate forecasting of travel time can not only provide better travel planning and route selection for passengers,improve passenger satisfaction,but also develops reasonable operational plans for the metro operation departments to improve operational efficiency.Therefore,it is of great significance to study the travel time prediction of subway passengers.The dataset of the correspding travel time based on the data of passengers entering and leaving the station of Fuzhou Metro Line 1 is built.Then,the machine learning algorithms are used to predict the travel time in this thesis.The main research work of this thesis is organized as follow:(1)Abnormal travel data processing for subway passengers.First,preprocessing the initial data and eliminating some obvious abnormal data.Then,according to the four statistics of travel time,travel time difference,travel time ratio and travel speed,combined with probability density and clustering algorithm,the subway passenger abnormal travel data is further eliminated,reducing the impact of anomalous data on subsequent experiments.The experimental results show that the abnormal data accounted for about 1.7%.Among them,the abnormality of the card is about 0.53%,and the abnormality of the one-way ticket is about 1.17%.(2)Forecast based on average travel time.It is mainly divided into two parts,the first part is based on historical data for one year,in which the average model,autoregressive integral moving average model,the support vector machine regression model and the long-term and short-term memory network are used to predict the average daily travel time in the future.The experimental results show that the support vector machine(SVM)regression model achieves the best prediction performance.The root mean square error and the mean absolute error obtained by the SVM are14.91 and 13.04 s,respectively.The second part is the time-based average travel time prediction,the support vector machine regression model has good prediction accuracy.Using real-time data prediction,the average absolute percentage error is about 3.65%,and the average absolute error is about 40.92s;Real-time data combined with historical statistics,the prediction is better,the average absolute percentage error is about 0.7%,and the average absolute error is about 8.01 s.(3)Prediction based on OD(Origin-Destination)travel time,that is,the prediction of travel time from the starting station to the terminal.Two basic prediction models based on expert experience and average model are proposed,and compared with multiple linear regression model,support vector machine regression model and error back propagation model,the final experimental results show that the prediction performance of multiple linear regression model is best.The average absolute error is104.52 s,which is nearly 23 s lower than other models,while the average relative percentage error is 10.17%,which is 2.42% lower than the best model of the other four models.