Abstract: Scientific and accurate short-term forecast of railway passenger traffic is the key to improve the competitiveness and service level of the railway passenger transport system. According to the characteristics of short-term railway passenger traffic, a variable weight combination forecasting model based on grey theory is proposed. In order to obtain the weight coefficient of different models at different times, the dynamic weights are tracked and predicted by using the generalized regression neural network. Taking the railway passenger traffic volume of 1~12 months in 2014 as the research object, the even grey model, the discrete grey model, the grey Verhulst model and the variable weight combination forecasting model are established respectively. The results of case analysis show that the average relative error of the three single model are 17.14%, 12.94% and 16.99% respectively, while the variable weight combination model is 7.01%, The forecasting accuracy of the variable weight combination forecasting model is obviously higher than the single model.

Key words: railway passenger traffic volume, variable weight, grey theory, generalized regression neural network, combination forecasting

摘要： 科学、准确的铁路客运量短期预测是提高铁路客运系统竞争力与服务水平的关键。针对铁路短期客运量的特点，提出了一种基于灰色理论的变权重组合预测模型。为了获取不同模型在不同时刻的权重系数，采用广义回归神经网络对动态权重进行跟踪和预测。以2014年1~12月份的铁路客运量为研究对象，分别建立均值GM（1，1）模型、离散GM（1，1）模型、灰色Verhulst模型以及变权重组合预测模型。实例分析的结果表明，三个单一模型的平均相对误差分别为17.14%、16.99%和12.94%，而变权重组合模型为7.01%，变权重组合预测模型的预测精度明显高于单一模型。

关键词: 铁路客运量, 变权重, 灰色理论, 广义回归神经网络, 组合预测