Data-driven dynamic robust resource allocation: Application to efficient transportation
The transformation to smarter cities brings an array of emerging urbanization challenges. With the development of technologies such as sensor networks, storage devices, and cloud computing, we are able to collect, store, and analyze a large amount of data in real time. Modern cities have brought to life unprecedented opportunities and challenges for allocating limited resources in a data-driven way. Intelligent transportation system is one emerging research area, in which sensing data provides us opportunities for understanding spatial-temporal patterns of demand human and mobility. However, greedy or matching algorithms that only deal with known requests are far from efficient in the long run without considering demand information predicted based on data. In this dissertation, we develop a data-driven robust resource allocation framework to consider spatial-temporally correlated demand and demand uncertainties, motivated by the problem of efficient dispatching of taxi or autonomous vehicles. We first present a receding horizon control (RHC) framework to dispatch taxis towards predicted demand; this framework incorporates both information from historical record data and real-time GPS location and occupancy status data. It also allows us to allocate resource from a globally optimal perspective in a longer time period, besides the local level greedy or matching algorithm for assigning a passenger pick-up location of each vacant vehicle. The objectives include reducing both current and anticipated future total idle driving distance and matching spatial-temporal ratio between demand and supply for service quality. We then present a robust optimization method to consider spatial-temporally correlated demand model uncertainties that can be expressed in closed convex sets. Uncertainty sets of demand vectors are constructed from data based on theories in hypothesis testing, and the sets provide a desired probabilistic guarantee level for the performance of dispatch solutions. To minimize the average resource allocation cost under demand uncertainties, we develop a general data-driven dynamic distributionally robust resource allocation model. An efficient algorithm for building demand uncertainty sets that compatible with various demand prediction methods is developed. We prove equivalent computationally tractable forms of the robust and distributionally robust resource allocation problems using strong duality. The resource allocation problem aims to balance the demand-supply ratio at different nodes of the network with minimum balancing and re-balancing cost, with decision variables on the denominator that has not been covered by previous work. Trace-driven analysis with real taxi operational record data of San Francisco shows that the RHC framework reduces the average total idle distance of taxis by 52%, and evaluations with over 100GB of New York City taxi trip data show that robust and distributionally robust dispatch methods reduce the average total idle distance by 10% more compared with non-robust solutions. Besides increasing the service efficiency by reducing total idle driving distance, the resource allocation methods in this dissertation also reduce the demand-supply ratio mismatch error across the city.
Computer Engineering|Civil engineering|Electrical engineering
Miao, Fei, "Data-driven dynamic robust resource allocation: Application to efficient transportation" (2016). Dissertations available from ProQuest. AAI10190772.