Optimizing OD-based up-front discounting strategies for enroute ridepooling services

The technological progress in the recent decade has greatly facilitated the large-scale implementation of dynamic enroute ridepooling services, such as Uber Pool and DiDi Pinche. To sustain a profitable enroute ridepooling service, a well-designed discounting scheme is crucial. This paper focuses on the optimization of up-front discounting strategies for enroute ridepooling service, under which passengers are notified of origin–destination(OD)-based discount ratios together with estimated ride time before the start of their trips and enjoy the discounted prices no matter if they succeed or fail to get matched afterward. Assuming that ridepooling demand of each OD pair decreases with its price and the estimated waiting and ride time, we propose to optimize the discounting strategy of each OD pair through two methods. In the first method, the ridepooling price of each OD pair is optimized independently and adjusted day-to-day based on historical information; and in the second method, we optimize the prices of all OD pairs simultaneously, with the complex interactions among the expected ride and waiting times and the demand rates of all OD pairs being considered and captured by a system of nonlinear equations. The nonlinear and non-convex optimization problem of the second method is solved by two derivative-free algorithms: Bayesian optimization and classification-based optimization. Based on a 15*15 grid network with 30 OD pairs and the real road network of Haikou (China), we conduct simulation experiments to examine the efficiency of the two algorithms and the system performance under different discounting strategies derived from the two methods. It is found that in comparison with a uniform discounting strategy, OD-based discounting strategies generated by both methods can bring about 3.84% more profit to the platform. In comparison with the independently optimized discounting strategies generated by the first method, the system optimal discounting strategy generated by the second method can further improve the platform profit by 5.55% and 2.71% on average in our grid-network and real road network experiments.