OPTIMISATION FOR CAR SHARING (BLUESG)
Background
​
BlueSG is an electronic vehicle carsharing company in Singapore, supported by the Land Transport Authority and Economic Development Board. It currently owns 530 vehicles and 253 charging stations with 1,003 charging points island-wide. With more than 42,000 members since its launch in 2017, there has been 364,000 rentals to-date with over 6.5 million km driven
​
Why data science? BlueSG needs to move cars
For the optimization model, we analyse the issue regarding the locations that blueSG should move their EVs into.
​
BlueSG faces the problem of varying demand.
​
There is varying demand for (i) different rental stations, (ii) different weekdays and at (iii) different times of the day. On weekdays, cars are generally rented out in the morning from housing estates to the CBD and downtown areas, thus it may be difficult to rent a car from housing areas in the morning. On the other hand, in the evening, the reverse happens and it may be difficult to find suitable parking in the housing estates. Thus, there might be profit loss and customer dissatisfaction when a customer has signed up as a member but cannot find available cars to rent.
To resolve this issue, blueSG can hire workers to move their electronic vehicles (EVs) between the different stations on different weekdays and times of the day.
Model Formulation
To formulate the model, we assume data is collected over 1 week, following 30 EVs at 8 stations (labelled 0 to 7). There are 468 data points. The following is the estimated average trip duration (in minutes) between any of the 8 stations. When the origin and destination is the same station (meaning the customer picked up and dropped off the car at the same location), the trip duration is estimated to be 25 min. Hence, a travel time matrix can be made as shown below. In addition, a 7 by 8 by 8 by 24 demand matrix is also generated, to represent: 7 (weekdays) x 8 (origin station) x 8 (destination station) x 24 (hours of the day).
​
There are 4 sets
There are 4 decision variables
Objective function
Constraints
Model Results
​
The optimal solution will result in a profit of $3615 per week and $516.53 per day, after accounting for the charging constraints. From the figures below, we can see that many cars are moved out of station 4, thus there might be excess cars there. Most cars are moved to stations 1 & 2, thus those areas are lacking cars. Overall, there is high demand at stations 3 and 4. Most cars have to be moved during the first and last hours of the day.
Extension
​
To see how well the model handles uncertainty in demand, demand was adjusted to randomly fluctuate from 60% to 140%. The number of cars that can be moved per hour is changed to 50 to meet the uncertainty. With demand uncertainty, profit can range between $3590 and $3665.
Since there is a larger fluctuation in profit, this could suggest spare capacity in blueCar’s fleet and blueCar can employ more marketing campaigns to increase its demand. In addition, demand can be boosted especially on Thursday and Wednesday, which has lower demand.
​
View Github code: link