This paper presents a multi-objective memetic algorithm based on request prediction for route planning in dynamic pickup-and-delivery problems. Historical data are used to predict the occurrence of new dynamic requests, based on which predictive routes are planned and tuned subsequently as the real requests occur. Two objectives namely route length and response time are optimized using multi-objective memetic algorithm that is a synergy of multi-objective genetic algorithm and a locality-sensitive hashing based local search. The proposed algorithm is tested on three benchmark problems and the experimental results demonstrate the efficiency of the algorithm.
