零担物流时序预测的SARIMA-GRU-BPNN组合模型及应用

Combined SARIMA-GRU-BPNN Model for LTL Logistics Time Series Prediction and Application

针对零担物流的需求物流量显著季节性、非线性和随机性特征使其预测难度大的问题,提出一种零担物流时序预测的SARIMA-GRU-BPNN组合模型的预测方法。使用季节性分解模型将物流量分解为趋势、季节性及残差,对趋势分量采用季节性差分自回归移动平均模型(SARIMA)拟合线性变化,对季节性分量采用门控循环神经网络(GRU)拟合季节性变化,对残差分量采用反向传播神经网络(BPNN)拟合非线性及随机性变化,组合重构得到最终预测值。实验结果表明,与自身单一模型SARIMA、GRU及BPNN相比,均方根误差(RMSE)分别降低31.5%、34.5%及47.1%;与其他单一模型灰色模型、支持向量机、长短期记忆网络及多元线性回归相比,RMSE分别降低71.3%、68.9%、54.4%及70.7%;与组合模型ARIMA-GRU、ARIMA-BPNN及ARIMA-SVM相比,RMSE分别降低31.0%、43.0%及56.1%,且趋势和季节性分量预测模型拟合优度达到92%和99%,有效降低整体预测误差,提升了预测精度和模型稳健性。

Aiming at the problem that the significant seasonal,nonlinear,and stochastic characteristics of the demanded material flow of less-than-truck-load logistics(LTL)make it difficult to predict,a prediction method of SARIMA-GRUBPNN combined model for LTL time series prediction is proposed.The seasonal decomposition model is used to decompose the logistics flow into trend,seasonality,and residual,the seasonal difference autoregressive moving average model(SARIMA)is used to fit the linear change for the trend component,the gated recurrent neural network(GRU)is used to fit the seasonal change for the seasonal component,and the back-propagation neural network(BPNN)is used to fit the nonlinear and stochastic change for the residual component,and the combination reconstruction is used to get the final prediction value.Based on the experimental results,the root mean square error(RMSE)is decreased by 31.5%,34.5%,and 47.1%when compared to single self-contained models SARIMA,GRU,and BPNN,respectively.Additionally,the RMSE is reduced by 71.3%,68.9%,54.4%,and 70.7%when compared to other single models gray model,support vector machines,long and short-term memory networks,and multiple linear regression,respectively.Furthermore,the RMSE is reduced by 71.3%,68.9%,and 54.4%when compared to combined models gray model,support vector machines,and long and short-term memory networks,respectively.In comparison to combined models ARIMA-GRU,ARIMA-BPNN,and ARIMA-SVM,the RMSE is reduced by 31.0%,43.0%,and 56.1%,respectively.The goodness-of-fit of the trend and seasonal component prediction models reaches 92%and 99%,effectively reducing the overall prediction error and improving prediction accuracy and model robustness.