earticle

영어

This paper designs a cluster-based electric vehicle relocation scheme for carsharing systems, aiming at generating a relocation schedule within a reasonable time bound by decomposing a large problem into several smaller ones. In order to develop a genetic algorithm for clustering, a feasible plan is encoded to an integer-valued vector in which intermediary stations locate at fixed positions and negative numbers separate clusters. The vehicles in overflow clusters are moved to underflow clusters through the intermediary stations first and then finally to underflow stations. The fitness function calculates the distance of all inter-station pairs in each cluster and selects the largest of them. Genetic operators continuously reduce the cost generation by generation. The performance measurement result, obtained by a prototype implementation, shows that the proposed clustering scheme linearly increases the cost according to the addition of a station, even if it is expected to increase exponentially. Moreover, the clustering plan converges to a stable cost in the early stages of the genetic evolution. These results indicate that we can overcome the stock imbalance problem and improve the service ratio.