Data mining-based dispatching system for solving the local pickup and delivery problem

The Local Pickup and Delivery Problem (LPDP) has drawn much attention, and optimization models and algorithms have been developed to address this problem. However, for real world applications, the large-scale and dynamic nature of the problem causes difficulties in getting good solutions within acceptable time through standard optimization approaches. Meanwhile, actual dispatching solutions made by field experts in transportation companies contain embedded dispatching rules. This paper introduces a Data Mining-based Dispatching System (DMDS) to first learn dispatching rules from historical data and then generate dispatch solutions, which are shown to be as good as those generated by expert dispatchers in the intermodal freight industry. Three additional benefits of DMDS are: (1) it provides a simulation platform for strategic decision making and analysis; (2) the learned dispatching rules are valuable to combine with an optimization algorithm to improve the solution quality for LPDPs; (3) by adding optimized solutions to the training data, DMDS is capable to generate better-than-actuals solutions very quickly.