While ride-hailing services only became a travel option in urban areas within the past few years, they are continuing to increase in number, size and popularity, especially as urban populations continue to grow. In 2019, the two largest markets for ride-hailing services in the U.S. and China saw their urban populations grow to 82% and 58% respectively. Much like Uber and Lyft in the U.S., in China, transportation company DiDi is the leading vehicle ride-share company.
Headquartered in Beijing, DiDi provides app-based transportation services, including taxi hailing, private car hailing, social ride-sharing and bike-sharing; on-demand delivery services; and automobile services, including sales, leasing, financing, maintenance, fleet operation, electric vehicle charging and co-development of vehicles with automakers. Throughout the year, DiDi provides services to more than 550 million passengers, resulting in over 10 billion rides.
Competing for the Wagner Prizes places you and your author team among the upper echelons of those within the O.R. and Analytics fields today.
The complexities of the ride-share marketplace – which is two-sided involving both the driver and passenger – must take into account multiple variables and factors. For the drivers, there is driver repositioning, order dispatching and pricing, and balancing driver income and time spent idle. For passengers, pickup distance, order response and fulfillment rates must be taken into consideration, as well as order-dispatching and matching.
The challenge of matching drivers and passengers is also a dynamic problem, with requests for rides streaming in throughout the day, and driver availability fluctuating as drivers log on and off, resulting in many matching decisions being made on the fly. In an effort to improve the overall marketplace efficiency, DiDi launched a project to improve their online dispatching and matching policy to increase total driver income, provide higher fulfillment and response rates, and shorten pickup distances.
The solution developed by DiDi is based on reinforcement learning and generalized policy iterations, is built on data collected and evaluated from its current dispatching system. Currently implemented in 20 Chinese cities, DiDi has seen an improvement of up to a 2% increase in auto response rate, fulfillment rate and total drive income, leading to improved travel experience for hundreds of millions of passengers and marking the very first successful industrial application of a solution of its kind in ride-sharing, for which DiDi was recognized with the 2019 INFORMS Daniel H. Wagner Prize.