交通运输工程与信息学报

公交客流是公交规划和运营调度的基础。针对短期公交客流的非线性、随机性和复杂性及支持向量机单核核函数自适应能力较弱的特点,提出一种基于多核最小二乘支持向量机的公交客流预测方法。该方法既考虑到了公交客流的历史数据规律,又顾及到短期公交客流的时变特性,充分利用了相关参数的知识信息。为了保证模型的自适应能力和提高模型的泛化能力,作者提出了综合评价指标,并采用改进遗传算法实现向量机参数优化。最后,结合LS-SVM工具箱,在MATLAB平台上实现长春市短期公交客流的预测。预测结果表明,提出的多核预测方法具有较高的准确性、较强的鲁棒性和自适应能力,在公交客流预测中有具有较好的应用价值。

Public transportation flow is the basic data for the public transport planning and operation scheduling.Based on the multiple-kernel least square support vector machine(MLS-SVM),the paper presented a new pubic transportation flow prediction model according to the non-linear,stochastic and complex characteristics of short-term public traffic flow.The proposed model not only considered the history data,but also took the character of the short-term public flow into account.In order to improve the suitability of the tradition model,a new evaluation index was proposed to portray the training performance of MLS-SVM.Crossover and mutation was modified with the genetic algorithm(GA),then using the improved GA optimized the penalty parameter and nuclear parameter.The model was applied to Chang-chun city,the result showed that the proposed model had satisfactory perform acne and robustness,and had good potential for predicting the short-term public transportation flow.