Objective At present, the rail transit systems in many Chinese cities are at a critical stage of rapid network expansion. On the one hand, the total volume, structure, and characteristics of rail transit network passenger flow are continuously evolving; on the other hand, the historical passenger flow data under the same network configuration are relatively limited, which brings new difficulties and challenges to passenger flow forecasting and analysis. Therefore, it is necessary to conduct research on the inbound/outbound passenger flow forecasting methods for urban rail transit stations under changing network conditions.

Method Based on multi-source correlation data analysis, the influencing factors of passenger flow forecasting are systematically extracted from two dimensions, i.e. the same network condition and the changed network condition. Considering the actual rapid changes of rail transit network structure, historical characteristic days of passenger flow patterns are first identified. Then, a network change coefficient and a natural growth coefficient are introduced to perform weighted corrections on passenger flows, and the exponential smoothing method is adopted to synthesize the multi-day corrected passenger flows. Based on the proposed forecasting method that takes network changes into account, a case study is carried out on Nanchang Metro.

Result & Conclusion  Under the condition of rapid network changes, the proposed method can effectively utilize historical data resources from different network stages, overcome the limitation of insufficient historical data under the same network configuration, thus resolving the difficulty that large-scale data-driven learning methods failing for network passenger flow forecasting. The forecasting errors for both inbound and outbound passenger volumes are within 9%. Compared with traditional forecasting methods, the proposed one exhibits higher accuracy and better applicability.