基于GRU深度网络的铁路短期货运量预测

GRU Deep Neural Network Based Short-term Railway Freight Demand Forecasting

铁路短期货运需求的预测是制定货运计划和进行运输组织的基础和重要条件。短期(月/日)货运量具有高度非线性、不确定性和序列依赖性的特征,难以预测。深度神经网络对时序数据有强大的学习能力,能较好拟合短期货运量的非线性特性。研究结构简单且具有高效记忆功能的门控循环单元(Gated Recurrent Unit,GRU)深度网络,分别建立单步和多步预测模型。最后应用某铁路局2011—2017年的货运量数据验证模型预测效果。结果表明,GRU单步模型对月度货运量预测的准确率达到98.87%,对日货运量预测的准确率达到94.56%,相比支持向量机回归、DP神经网络和LSTM等模型在准确率和均方根误差指标上更优、效果更好。

Short-term railway freight demand forecasting is extremely important for formulating freight transportation plans and strengthening transportation organization.Short-term(monthly/daily)freight demand,featuring high nonlinearity,uncertainty and sequence dependence,makes it difficult to predict.The strong learning ability of the deep neural network for time series data can better fit the nonlinear characteristics of short-term freight demand.In this paper,the Gated Recurrent Unit(GRU)deep neural network with simple structure and efficient memory function was studied to establish single-step and multi-step prediction models respectively.Finally,the freight demand data of a railway bureau from 2011 to 2017 was used to verify performance of the proposed methodology.The results show that the GRU multivariate single-step model has accuracy rates of 98.87%and 94.56%for the monthly and daily freight demand forecasting,respectively.Compared with support vector machine regression,DP neural network and LSTM models,the GRU neural network is more precise with lower root mean square error.