基于无监督学习的实时公交动态调度的研究

针对广州智能公交调度的优化问题，提出一种基于无监督学习的实时公交动态调度算法，结合乘客利益和公交公司利益总体最优为目标，通过无监督学习方法学习到公交客流出行特征表达的提取，利用吸引子传播(affinity propagation，AP)聚类算法的优化数据集与支持向量机(support vector machine，SVM)的训练样本集相结合建立预测模型训练，运用公交线网发车间隔和加权系数的目标函数优化调度数学模型，将多源信息融合及多策略的实时公交动态调度算法引入到求解模型中，利用深度学习的异常突发事件分类检测方法实现调度优化模型的实时调整。实验结果表明，AP聚类算法程序运行耗时16 s、高峰发车间隔5 min，比遗传算法运行效率更高、时间间隔更精确，实例证明模型和算法具有实用性和可靠性。

Research on real time bus dynamic scheduling based on unsupervised learning

Aiming at the optimization of intelligent bus dispatching in Guangzhou, a real-time dynamic bus dispatching algorithm based on unsupervised learning is proposed and applied to this problem. Combining the interests of passengers and bus companies as the goal, the expression of bus passenger outflow characteristics learned by unsupervised learning method is extracted, and the attraction is utilized. Affinity Propagation (AP) clustering algorithm optimization data set and support vector machine (SVM) training sample set are combined to establish prediction model training. The objective function of bus network departure interval and weighting coefficient is used to optimize scheduling mathematical model, and multi-source information fusion and multi-strategy are achieved. Real-time bus dynamic scheduling algorithm is introduced into the solution model, and the real-time adjustment of the scheduling optimization model is realized by deep learning abnormal incident classification detection method. The experimental results show that the AP clustering algorithm takes 16S to run, and the peak workshop interval is 5Min, which is more efficient and precise than genetic algorithm. An example shows that the model and algorithm are practical and reliable.