针对现有城市轨道交通短时客流量预测单一模型可能存在预测不稳定的问题，本文提出一种基于奇异谱分析(Singular Spectrum Analysis, SSA)和支持向量回归(SVR)相组合的预测模型。该组合模型利用奇异谱分析(SSA)将轨道交通原始时间序列客流数据进行分解和重构，对重构后的时间序列进行奇异值按从大到小进行排序，得到含有原始时间序列数据主要信息成分的重构序列，将重构后的时间序列作为支持向量回归模型(SVR)的输入，最后进行各站点的短时进站客流预测。本文采集2015年11月份北京市全网的城市轨道交通进站客流数据，对提出的短时客流预测模型进行验证和对比分析。结果表明，本文提出的组合模型预测精度相比ARIMA模型、SVR模型、CNN-LSTM模型和T-GCN模型具有更高的预测精度和更稳定的预测表现，具有一定的实际意义。

In view of the existing urban rail transit short-term traffic forecasting a single model may predict instability problems, this paper puts forward a kind of based on singular spectrum analysis (SSA), and by combining support vector regression (SVR) forecasting model. The combined model uses singular spectrum analysis (SSA) to decompose and reconstruct the original time series passenger flow data of rail transit, and sort the reconstructed time series by singular values from large to small to obtain the main information containing the original time series data. The reconstructed sequence of the components uses the reconstructed time series as the input of the support vector regression model (SVR), and finally the short-term inbound passenger flow prediction of each station is carried out. This paper collects the urban rail transit passenger flow data of the entire network in Beijing in November 2015, and validates and compares the proposed short-term passenger flow prediction model. The results show that the prediction accuracy of the combined model proposed in this paper is higher and more stable than ARIMA model, SVR model, CNN-LSTM model and T-GCN model, which has a certain practical significance.

国家自然科学基金(71864022)