关键词: 城市交通, 出行方式预测, 支持向量机, 粒子群算法, 参数选择

Abstract: Resident trip mode choice is a complex nonlinear problem which is affected by many factors.This paper models the resident trip mode with the support vector machine (SVM) method.It uses the classification accuracy as fitness function in the sense of cross validation and then adopts the particle swarm optimization (PSO) algorithm to select parameters.The choice randomness of specifying the parameters are avoided and the workload of the parameter selection are reduced.The empirical studies indicate that the PSO based parameters optimization in SVM is feasible.Compared with the BP neural network model, the SVM model achieves a better prediction performance for resident trip mode choice.The prediction accuracy of the SVM model is improved by 5% to BP neural network model.The SVM model obtains the forecasting accuracy rate of 86.20% to the model setting and 82.31% to the test samples.The empirical results show that the model can be used for resident trip mode forecasting and provides a practical guidance to urban traffic planning and travel demand forecasting.

Key words: urban traffic, trip mode prediction, support vector machine (SVM), particle swarm optimization (PSO), parameter selection