基于小波变换的RBFNN与LSSVM高铁客流量组合预测

Combined RBFNN and LSSVM for Forecasting High-speed Rail Passenger Flow Based on Wavelet Transform

由于受到诸多因素的影响,高铁客流量呈现出非线性、时变性等特征,单一预测法很难准确揭示这些复杂变化特征.为解决该问题,本文提出一种基于小波变换(wavelet)的径向基神经网络(RBFNN)与最小二乘支持向量机(LSSVM)高铁客流量组合预测方法.首先,先通过小波变换将高铁客流量数据序列分解为低频趋势分量序列与高频随机分量序列,再利用RBFNN和LSSVM分别预测这2种分量序列,最后将这2种分量序列的预测值叠加为最终组合预测值.以我国某高铁站客流量数据为例,验证wavele t-RBFN N-LSSVM组合预测法的有效性,结果表明,wavele t-RBFN N-LSSVM的RMSE、MAE、MPE、Theil、HRMSE、LLF值均小于RBFNN和LSSVM的对应值,且拟合优度R 2值为0.1118,高于RBFNN的0.0325和LSSVM的0.0070,相较于RBFNN和LSSVM的预测精度更高,适合于高铁客流量的短期预测.

The high-speed rail passenger flow shows nonlinear and time-varying due to the influence of many factors.It is difficult to accurately reveal these complex changing characteristics by a single prediction method.To solve this problem,this paper proposed a combination method integrating radial basis function neural network(RBFNN)and least squares support vector machine(LSSVM)based on wavelet transform for forecasting high-speed railway passenger flow.Firstly,the data sequence of high-speed railway passenger flow was decomposed into low-frequency trend and high-frequency random component sequence by using wavelet transform technology.Then,the two component sequences were predicted by RBFNN and LSSVM respectively.Finally,the predicted values of the two component sequences were integrated into the final predicted values.Taking passenger flow data of a high-speed railway station in China as an example,the paper verified the effectiveness of the combined forecasting method of wavelet RBFNN LSSVM.The results indicate that the RMSE,MAE,MPE,Theil,HRMSE and LLF values of wavelet RBFNN LSSVM are less than the corresponding values of RBFNN and LSSVM,and the goodness of fit R 2 value is 0.1118,which is higher than the R 2 value of RBFNN 0.0325 and the R 2 value of LSSVM 0.0070,which is more accurate than that of RBFNN and LSSVM,it is suitable for short-term prediction of passenger flow of high-speed railway.