Abstract: With the completion and operation of high-speed railway passenger dedicated lines in succession, a certain amount of passenger flow historical data has been accumulated. In order to reasonably allocate transport capacity and improve transport service quality, it is necessary to fully mine the historical data for short-term prediction of passenger flow. On the basis of considering the samples of different length time series, an integrated model for short-term passenger flow prediction was proposed. Firstly, based on the sampling of the original OD passenger flow data between stations, the variable length time series comprehensively reflecting the characteristics of passenger flow were obtained as the samples. Secondly, the wavelet decomposition and ARIMA model were combined to build a weak passenger flow prediction model. Finally, the AdaBoost integration algorithm was used to combine multiple weak models to build a strong passenger flow prediction model. Based on the passenger flow data of Chongqing-Wanzhou line, the parameters of the proposed model were calibrated and tested. The research shows that the proposed model has better prediction accuracy and generalization ability than GM(1,1) and ARIMA, and has an average improvement of 38.12%, 67.78% and 38.52% in the three indexes of average absolute error, average relative error and mean variance.

Key words: traffic and transportation engineering; time series; passenger flow forecast; wavelet decomposition; ARIMA model; Adaboost ensemble algorithm

摘要： 随着高铁客运专线陆续建成并投入使用，积累了一定的客流历史数据，为合理分配运能及提高运输服务质量，需充分挖掘历史数据对客流进行短期预测，在考虑不同长度时间序列样本的基础上提出客流短期预测集成模型。首先，基于站间OD原始客流数据进行抽样得到全面反映客流特征的变长时间序列作为样本；其次，将小波分解和ARIMA模型结合构建客流预测弱模型；最后，使用Adaboost集成算法思想将多个弱模型组合起来构建客流预测强模型，并以重庆渝万线客流数据为基础对模型进行参数标定与检验。研究表明：提出的模型对比GM(1,1)和ARIMA模型有较好的预测精度和泛化能力，在平均绝对误差、平均相对误差和均方差这3个指标上平均有38.12%，67.78%和38.52%的提高。

关键词: 交通运输工程；时间序列；客流预测；小波分解；ARIMA模型；Adaboost集成算法