Short-Term Passenger Flow Forecast At The Initial Stage Of New Urban Rail Transit Line

With the expansion of the planning and construction scale of the urban rail network in recent years,the construction of new lines has continued to maintain a high speed development trend.The initial stage of new line is a critical period for cultivating and attracting passenger flow.The opening of new line stations will bring impact and influence to the existing line network under the conditions of network formation.The short-term passenger flow forecast can provide data support for the development of new line operation plans and capacity resource allocation to ensure the efficiency of new line operation and improve service levels.Therefore,the initial stage of new line is an important scenario in urban rail transit passenger flow forecasting,which is particularly crucial for optimising urban rail transit organisation.The paper firstly analysed the passenger flow characteristics of new urban rail transit lines at the initial stage of operation from both temporal and spatial dimensions.In terms of temporal distribution,the growth trend of daily passenger flow was analysed,as well as the characteristics of changes in the short-term distribution of passenger flow.In terms of spatial distribution,the differences in passenger flow characteristics of different station attributes and locations were analysed.The paper examined the passenger flow generation mechanism of the new line and discussed the factors influencing passenger flow in the initial stage of the new line,which provides a reference for the classification of the passenger flow development stage and passenger flow forecast of the new line.According to the pattern of passenger flow during the initial stage of a new line,the development stages of a new line were divided into the incubation period,the growth period and the maturity period.When a new line opens,it enters the incubation period with high passenger growth.When there is negative growth at some stations,it indicates a slowdown in passenger growth.The new line enters the growth period with strong passenger fluctuation and significant noise.When the change in passenger flow is less than 10% for two consecutive weeks,it indicates that the range of passenger change is the same as that of the mature line.During this time,the change in passenger flow becomes stable,and the new line enters the maturity period.By dividing the stages of development,the key points and difficulties of passenger flow forecasting at each stage were identified in the paper.Further,a framework for passenger flow forecasting at the initial stage is constructed.In the incubation period,the point of interests(POI)statistics were quantified as indicators of the scale of the sites around the stations.A hierarchical clustering method was used to find existing stations that are similar to the new line stations,so as to compensate for the lack of historical data for the new line.A K-nearest neighbor(KNN)model was used to predict the passenger flow of the new line by constructing a state vector based on the scale of the site.Depending on the confidence level of the input passenger flow data,quadratic exponential smoothing was used to improve the utilisation of the historical passenger flow of the new line and to correct the prediction results.A combined ARIMA-Kalman forecasting model was constructed for the growth period.The model separates the passenger flow characteristics into daily passenger flows and timesharing coefficients.The ARIMA model with natural logarithm transformation was used to fit the trend of daily passenger flow from growth to saturation.Based on the variation pattern of passenger flow noise and the difference of noise amplitude range at different moments,the measurement noise covariance matrix was optimised to enhance the extraction of effective noise information.A time-sharing coefficient prediction model based on improved Kalman filter was proposed.Finally,the daily passenger flow and passenger flow time-sharing coefficient characteristics were fused to obtain the growth period passenger flow.The Nanjing Metro Line 4,which opened in 2017,was used as an example for validation.The results showed that the KNN model is effective in predicting passenger flow during the incubation period of the new line and the error converges quickly with time.The prediction accuracy of KNN is 86.19% within one week of opening.The combined ARIMA-Kalman model for the growth period has a higher prediction accuracy,which is more accurate than neural networks and other prediction models under the same conditions.The average accuracy of the combined ARIMA-Kalman model is 88.63%.The error decreases faster,and the accuracy reaches 91.21% in the seventh week of opening,which improves the current situation that the short-term passenger flow prediction of new lines is limited by small samples and high noise data.Besides,the model has good robustness and can better adapt to the dynamic fluctuation of passenger flow and frequent capacity adjustment of the new line.