关键词: 混合核支持向量回归机（SVM）, 参数优化, 黄金分割, 混沌粒子群优化, 站点客流量

Abstract: Real-time traffic management of subway operation scheduling departments depends on subway station passenger flow. This paper builds a hybrid kernel Support Vector regression Machine （SVM） forecasting model with a linear combination of radial basis function and polynomial kernel function. This paper proposes the Golden section Chaotic Particle Swarm Optimization （GCPSO） to search the optimal combination of mixed kernel SVM parameters. The results of short-term passenger flow prediction with the hybrid kernel SVM in Guangzhou Metro Line 3 indicate that GCPSO mixed kernel SVM forecasting model is superior to other three SVM prediction methods and Elman neural network forecasting with high precision of short-term passenger flow forecasting, good fitting of predicted and measured data and less proportional errors.

Key words: hybrid nuclear Support Vector regression Machine（SVM）, parameter optimization, golden section, chaotic particle swarm optimization, passenger flow