Managing how and when to charge the batteries of shareable electric scooters is one of the biggest logistic challenges of any new and old firm in the industry.
Most of the players were used to pick up the scooters that had run out of battery at night to bring them to their facilities, recharge them, and redistributing them in the early morning. Many different - and disruptive - strategies are being tested to reduce the related costs without affecting the level of service, but in this particular case, the task was to optimize the traditional process of "pick-up-recharge-redistribution" with more efficient scheduling.
Most of the players were used to pick up the scooters that had run out of battery at night to bring them to their facilities, recharge them, and redistributing them in the early morning. Many different - and disruptive - strategies are being tested to reduce the related costs without affecting the level of service, but in this particular case, the task was to optimize the traditional process of "pick-up-recharge-redistribution" with more efficient scheduling.
Modeling the actual distribution of the scooters' usage I was able to create new scheduling that not only allows a more constant level of service but also drastically reduces the required warehouse dimensions (-55%), number of trucks (-53%), and employees (-11%) increasing, on the other hand, the number of total trips during the day (+45%).
The starting dataset consisted of over 22'000 observations across 4 cities representing the bounties' appearance (a bounty is created whenever a scooter runs out of battery). Not having access to any proprietary tool of the firm and not knowing its internal coding knowledge, the analysis has been conducted in Excel.
The model considers the necessary time to pick up the scooters and bring them to the warehouse, the time to recharge, and the time to bring the scooters back in position. Assuming for every hour of the day, a normal distribution of the bounties, a Monte Carlo Simulation allows a sensitivity analysis to test the robustness of the model itself.
The target function to be minimized considers all the costs related to the logistic.
As customizable constraints and variables there are:
- Max available trucks
- Initial number of scooters in the city
- Initial number of scooters in the warehouse
- Minimum level of service (minimum number of scooters in the city charged)
- Maximum size of the warehouse
- Working shifts
- Various costs (staff, truck, charging, holding)
- Max available trucks
- Initial number of scooters in the city
- Initial number of scooters in the warehouse
- Minimum level of service (minimum number of scooters in the city charged)
- Maximum size of the warehouse
- Working shifts
- Various costs (staff, truck, charging, holding)
Even though the model provides useful data for decision-making, some assumptions and limitations are:
- Collection and redistribution cannot be performed at the same time by the same truck
- The timing of the bounties is related to the actual logistic strategy, more information about the scooters' utilization are needed; furthermore, the sample is too small to properly assume a normal distribution
- Traffic and rush hours have not been considered
- No "Travelling Salesman Problem" has been considered: a given, average time for collection has been used for every scooter
- Collection and redistribution cannot be performed at the same time by the same truck
- The timing of the bounties is related to the actual logistic strategy, more information about the scooters' utilization are needed; furthermore, the sample is too small to properly assume a normal distribution
- Traffic and rush hours have not been considered
- No "Travelling Salesman Problem" has been considered: a given, average time for collection has been used for every scooter
A sample of the model with modified data to protect the information of the firm is retrievable on my GitHub profile as "Scooter Logistic Simulation.xlsx".
