基于小波分析的最小二乘支持向量机轨道交通客流预测方法

A Wavelet Analysis Based LS-SVM Rail Transit Passenger Flow Prediction Method

针对城市轨道交通客流预测问题,采用离散一维Daub4小波分析方法对某一时间段的原始客流时间序列数据进行分解;以分解得到的高频分量和低频分量为样本数据,对最小二乘支持向量机进行训练,确定最小二乘支持向量机的核参数σ,以及系数α和b。利用训练后的最小二乘支持向量机预测未来一段时间客流时间序列数据的高频分量和低频分量,然后再利用Daub4小波分析方法对预测的高频分量和低频分量进行数据重构,从而得到预测的未来一段时间客流时间序列数据。与历史平均预测法和灰色预测法进行比较,结果表明,基于小波分析的支持向量机客流预测方法用于轨道交通短期客流预测具有更好的精度。

To deal with rail transit passenger flow forecasting problem, the discrete one-dimensional Daub4 wavelet analysis method was adopted to decompose the original time series data of passenger flow in a cer- tain period into different low-frequency and high-frequency components, which were used as sample data to train least squares support vector machine （LS-SVM） to determine LS-SVM nuclear parameter a, coeffi- cients a and b. The trained LS-SVM was used to predict the low-frequency and high-frequency components of passenger flow time series data in a future period of time. Then Daub4 wavelet analysis method was a- gain adopted to reconstruct the predicted low-frequency and high-frequency components to obtain the pre- dicted time series data of passenger flow in a future period of time. Compared with historical average pre- diction method and gray prediction method, results show that wavelet analysis based SVM passenger flow prediction method has higher accuracy in short-term passenger flow prediction for rail transit.