Application of genetic algorithm in vehicle routing problem with stochastic demands

The paper considered a version of vehicle routing problem where customers' demands were stochastic, and was applied in traffic and communication, manufacture and control, integrate circuit design extensively. Assuming that actual demand was revealed only upon arrival of a vehicle at the location of each customer, and the demand could not be partitioned, the goal consisted of minimizing the expected distance traveled in order to meet all customers' demands. Firstly, two prior policies, tour policy and multi-tour policy were provided, and their asymptotic properties were analyzed. To find the prior solution of tour policy and multi-tour policy, genetic algorithms with different neighborhood structures were designed. Experiments demonstrated validity of tour policy and multi-tour policy, and showed superiority of genetic algorithm with combined neighborhood.