车辆合乘问题的两阶段分布式估计算法

针对智慧交通中多车辆合乘问题，提出一种分布式并行计算环境下的合乘模型. 利用合乘概率矩阵的先验知识，实现更高效的运算和求解.当合乘概率矩阵不是单位 矩阵时，合乘模型被增广为车主合乘和乘客合乘两个阶段.两阶段分布式估计算法运用可行合乘解的合乘概率矩阵，作为一种随机优化方法求解最优值.根据可搭乘矩阵初始化合 乘概率矩阵，并在优化过程中连续更新合乘概率矩阵.车主同乘客分离优化，减少了出行车辆，并实现了互相搭乘的合乘模型.通过合乘模型的优化迭代能够为乘客挖掘出高效可 行的搭乘路线.实验结果表明，该合乘模型具有平均等待时间少、平均载客量大、人均行驶 距离短的高效出行特点.

Two-stage Estimation of Distribution Algorithm to Solve Multi-vehicle Carpooling Problem

A multi-carpooling model is proposed for the multi-vehicle carpooling problem in a distributed parallel computing environment. The prior knowledge of the carpooling probabilistic matrix is used for more efficient computing and effective solutions. When the carpooling probabilistic matrix is not the identity matrix, the multi- carpooling model is augmented into two stages of drivers’ridesharing and passengers’ ridesharing. A two-stage estimation of distribution algorithm is proposed as a stochastic optimization method to solve the optimum with a carpooling probabilistic matrix of promising carpooling solutions. A ridable matrix initiates the carpooling probabilistic matrix, and the optimization consists of a series of incremental updates of the carpooling probabilistic matrix. The optimization process of drivers and passengers is separated; hence, the carpooling model implements the mutual ridesharing to decrease vehicles demanded. The carpooling model mines efficient and compromised ridesharing routes for shared riders by the optimization iterations. Experimental results indicate that the carpooling model has the characteristics of effective and efficient traffic including shorter waiting time, more passenger load, and less average riding distance.