Research On Short-time Origin-Destination Passenger Flow Forecasting For Urban Rail Transit Network

In recent years,the scale of urban rail transit construction has been continuously improved,and the pressure of citizen travel has been greatly relieved,but there are still some potential problems,such as congestion during peak hours,mismatch between vehicle supply and passenger demand,hidden safety hazards in which a large number of commuter passenger flows converge in underground spaces,the local passenger flow surge needs rapid evacuation under special operating environments such as large-scale events,bad weather,and subway failures,and other issues.To solve the above problems,it is urgent to grasp the short-term change law of Origin to Destination(OD)passenger flow of rail transit,and the automatic ticketing system of urban rail transit can collect a large amount of data information generated by citizens’ daily travel in real time,making it possible.In this paper,the urban rail transit line network dynamic OD passenger flow multimodel ensemble prediction model is built by referring to the idea of ensemble learning,and the short-term change rules of the rail transit OD passenger flow is predicted to solve the above problems.Firstly,data preprocessing is performed on AFC,stations,lines,ticket data,and external weather data,and multiple single data sets are integrated into multiple sources to form an organic whole,which provides data support for the next analysis of passenger flow influencing factors;Secondly,the OD passenger flow is analyzed from the type,periodicity and volatility,K-Means clustering is used to classify the rail transit OD pairs and include it in the influencing factors,and then this paper analyzed and summarized the influencing factors of short-term OD passenger flow from three aspects of time,space and external weather,which provides input for the next model prediction.Thirdly,four models with different functions such as MLR,KNN,XGBoost and LSTM are selected as sub-model,and use particle swarm optimization to optimize the hyper-parameters of the sub-model,and build two multi-model ensemble passenger flow prediction models based on regression and classification by referring to the idea of ensemble learning.Finally,the built ensemble models are used to make a short-term prediction of the OD passenger flow of the normalized and non-normalized for rail transit line network.Under normalization,the prediction result of average over the same period in history is used as the baseline to compare with the constructed model in terms of overall,OD pair type,working day and non-working day,peak and non-peak,etc.Through the comparison of model performance,it is found that the prediction effect of the ensemble model is better than that of the single sub-model.Among them,the classification and integration model had the best effect,and the root mean square error can be reduced by up to 20% compared with average over the same period in history under the scene with greater randomness,which verifies the superiority of the proposed model.Under normal conditions,the model is applied through the Beijing rail transit failure case.In the case where averaging over the same period in history no longer works,the prediction effect of the classification ensemble model is considerable,and the prediction effect is the best when the time granularity is 10 minutes,and then some suggestions on emergency strategies are given in three aspects: travel guidance and emergency response for bus connection.The research results of this paper can enrich the short-term passenger flow OD prediction model system of urban rail transit,provide more accurate input for the dynamic operation and control system of urban rail transit,make the formulation and adjustment of driving plans more reasonable.Furthermore,it can provide data support and scientific basis for effectively alleviating the mismatch between supply and demand,peak congestion,quickly formulating emergency plans,effectively evacuating people in conjunction with other transportation methods,and for collaborative optimization of the city’s comprehensive public transportation system.