基于VMD-LSSVM的扇区流量短期预测

Short Term Prediction of Sector Traffic Based on VMD-LSSVM

对扇区流量进行短期预测,是精准实施扇区流量优化和管理措施的前提。基于分解集成预测方法论,建立了变分模态分解-最小二乘支持向量机(Vibrational mode decomposition-least square support vector machines,VMD-LSSVM)预测模型。首先,应用变分模态分解(Vibrational mode decomposition,VMD)方法将扇区流量时序数据分解为若干个模态;然后,使用最小二乘支持向量机(Least square support vector machines,LSSVM)模型分别对模态进行预测;接着,对模态的预测结果进行加和集成,得到了最终的预测值。算例计算结果显示,针对60 min统计尺度流量时间序列,VMD-LSSVM模型在1~6 h的均等系数(Equal coefficient,EC)值为0.97,在7~12 h的EC值为0.94;与差分自回归滑动平均模型(Autoregressive integrated moving average model,ARIMA),反向传播(Back propagation,BP)神经网络和LSSVM单一模型相比,VMD-LSSVM模型1~6 h的EC值分别提升了11.5%、7.8%、4.3%;与完整聚合经验模态分解(Compete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)-LSSVM、CEEMDAN-BP和VMD-BP相比,提升了2.1%、6.6%、5.4%;与30 min和15 min统计尺度相比,的EC值分别提升了6.6%和19.8%;针对时间普适性的8次实验,EC值均在0.94以上,针对27个扇区普适性的实验,有24个扇区的EC值在0.9以上。算例结果表明,VMD-LSSVM模型具备良好的预测性能和较好的普适性,用于扇区流量短期预测是可行的和有效的。

Short term prediction of sector traffic is the premise of accurately implementing sector traffic optimization and management measures.Based on the decomposition integration prediction methodology,a vibrational mode decomposition least square support vector machine(VMD-LSSVM)prediction model is established.Firstly,the VMD method is applied to decompose the traffic into several sectors.Then,the LSSVM model is used to predict the modes.The modal prediction results are added and integrated to obtain the final prediction value.The calculation results show that the prediction accuracy of the VMD-LSSVM model is 0.97 in 1—6 h and 0.94 in 7—12 h.Compared with the first mock exam model of autoregressive integrated moving average model(ARIMA),back propagation(BP)and LSSVM,the prediction accuracy of the VMD-LSSVM model 1—6 h increased by 11.5%,7.8%,4.3%,respectively,and 2.1%,6.6%,5.4%compared with compete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)-LSSVM,CEEMDAN-BP and VMD-BP,respectively.Compared with 30 min and 15 min statistical scales,the prediction accuracy is improved by 6.6%and 19.8%,respectively.For the eight experiments of time universality,the prediction accuracy is more than 0.94.For the experiments of 27 sectors,the prediction accuracy of 24 sectors is more than 0.9.The example results show that the VMD-LSSVM model has good prediction performance and good universality,and it is feasible and effective for short-term prediction of sector traffic.