《Table 2 Performance comparison of forecasting results with30min ahead》

  提示：宽带有限、当前游客访问压缩模式

本系列图表出处文件名：随高清版一同展现

《基于核自适应滤波的短时交通流量在线预测（英文）》

1. 获取 高清版本忘记账户？点击这里登录

1. 下载图表忘记账户？点击这里登录

The parameter settings are as follows.KRLS，KLMS，KELM，KPLS，KPCA，LSSVM and SVM all select Gaussian kernel functions with kernel function width ofσ=0.5；the regularization parameter of KELM isγ=200；the potential variables of KPLS is p=20；the learning rate of KLMS isη=0.3；the parameters of SVM and LSSVM are C=0.5，ε=0.3.In KPCA-SVM，KPCA-LSSVM and KPCA-KELM methods，the nonlinear principal components of KPCA method are 15；SVM selects linear kernel function；ALD-KRLS algorithm maximum dictionary capacity Mmax=500and thresholdυ=0.000 01.The learning rate of KLMS isη=0.3.FB-KRLS has the fixed memory size M=300，learning rateη=0.1，and regularization parameterλ=0.1.Tables 1and 2give the forecasting results with 15min and 30min in advance using different evaluation methods.