Data-driven optimization for drone delivery service planning with online demand
针对无人机配送中实时需求不确定的挑战，提出一种数据驱动优化方法，通过监督学习预测网络链路优先级并动态调整储备容量，以最大化累积利润，仿真表明优于短视策略。
In this study, we develop an innovative data-driven optimization approach to solve the drone delivery service planning problem with online demand. Drone-based logistics are expected to improve operations by enhancing flexibility and reducing congestion effects induced by last-mile deliveries. With rising digitalization and urbanization, however, logistics service providers are constantly grappling with the challenge of uncertain real-time demand. This study investigates the problem of planning drone delivery service through an urban air traffic network to fulfill dynamic and stochastic demand. Customer requests – if accepted – generate profit and are serviced by individual drone flights as per request origins, destinations and time windows. We cast this stochastic optimization problem as a Markov decision process . We present a novel data-driven optimization approach which generates predictive prescriptions of parameters of a surrogate optimization formulation. Our solution method consists of synthesizing training data via lookahead simulations to train a supervised machine learning model for predicting relative link priority based on the state of the network. This knowledge is then leveraged to selectively create weighted reserve capacity in the network and via a surrogate objective function that controls the trade-off between reserve capacity and profit maximization to maximize the cumulative profit earned. Using numerical experiments based on benchmarking transportation networks, the resulting data-driven optimization policy is shown to outperform a myopic policy. Sensitivity analyses on learning parameters reveal insights into the design of efficient policies for drone delivery service planning with online demand.