Com-DARP benchmark.zip

An instance characteristics are read as in the following example : "100_3_1.3" means 100 users, 3rd instance of the benchmark with 100 users, maximum ride time coefficient set to 1.3. In total, we generated 24 instances: 8 instances with 100 users, 8 instances 200 users and 8 instances with 400 users. <br> A user has two requests: one morning request and one afternoon request. The pickups and drop-offs locations are randomly distributed on a (-10, 10) grid. For a user, the morning pickup location corresponds to the afternoon drop-off location and the morning drop-off location corresponds to the afternoon pickup location. The morning and the afternoon loads are both equal to 1 and the service duration is always equal to 3 minutes at pickup and drop-off. A user maximum ride time is set to its direct ride times multiplied by a coefficient (1.2, 1.3, 1.4, 1.5). A request maximum ride time is set to the user maximum ride time minus the pair request direct ride time. Additionally, we raise to 15 minutes any maximum ride time lower than this value. The time windows for the morning drop-off and afternoon pickup are set first. The maximum time window for the morning drop-off is randomly selected according to a normal distribution centered on 8am with a variance of 90 minutes. Similarly, the minimum time window for the afternoon pickup is randomly selected according to a normal distribution centered on 4pm with a variance of 90 minutes. Time windows are opened for 10 minutes. Then, the morning pickup and the afternoon drop-off time windows are set according to the time window tightening procedure described in Cordeau (2006). There is one depot located in the center of the map. The service time is equal to zero minute at the depot. A fleet of infinite homogeneous vehicles leaves and returns to this depot. Each vehicle has a capacity of 6 passengers, cost 0.25€ per kilometer, 0.5€ per minute and has a fixed cost of 50€ if used during the day. <br> <br>

实例特征可参照如下示例解读：形如“100_3_1.3”的实例代表包含100名用户的基准测试集的第3个实例，且其最大乘车时间系数设为1.3。本次共生成24个实例：100用户规模的实例共8个，200用户规模的实例共8个，400用户规模的实例共8个。

每位用户对应两项出行请求：一项为早间请求，另一项为下午请求。所有上下车地点均随机分布于(-10, 10)的网格平面内。对于每位用户而言，早间上车地点与下午下车地点重合，早间下车地点与下午上车地点重合。早间与下午的出行载荷均为1，且上下车环节的服务时长固定为3分钟。用户的最大乘车时长为其直达乘车时长乘以系数（1.2、1.3、1.4、1.5）。单趟请求的最大乘车时长则为用户的最大乘车时长减去该请求配对的另一趟请求的直达乘车时长。此外，所有小于15分钟的最大乘车时长均被调整为15分钟。

首先设定早间下车与下午上车的时间窗。早间下车的最晚时间窗中心设定为上午8点，方差为90分钟，其具体取值通过正态分布随机生成。同理，下午上车的最早时间窗中心设定为下午4点，方差为90分钟，同样通过正态分布随机生成。所有时间窗的开放时长均为10分钟。随后，早间上车与下午下车的时间窗将参照Cordeau (2006)提出的时间窗收紧流程进行设定。

地图中心设有一个配送场站（depot），在该场站的服务时长为0分钟。本次实验采用无限规模的同质车辆车队，所有车辆均从该配送场站出发并最终返回。每辆车的载客容量为6名乘客，运营成本为每公里0.25欧元、每分钟0.5欧元，若当日启用该车辆，则需支付50欧元的固定成本。