For the LSCARP, the scale of the problem and solution space becomes very large. RBD-MAENS adopts the divide and conquer strategy to decompose the problem into many sub-problems, and then it combines CC framework and local search to improve the quality of solution. In RBD-MAENS, the best route grouping information so far is used to decompose the whole problem and then solve each sub-problem combing with CC framework. The performance of the cooperative evolutionary framework is largely determined by the route grouping scheme. In the process of route grouping, it firstly defines the distance between two nodes, which is routing distance instead of the geographic information in the Euclidean distance between vertices. Then the Dijkstra algorithm is used to determine the shortest distance between two arbitrary nodes. Finally, the distance between two routes based on the distance between nodes in these two routes is determined. Based on this definition, the grouping becomes the clustering of the routes. The smaller the distance between the two routes is, more likely they…show more content… The smaller the value of J(C) is, the higher quality clustering has. If the clustering quality after the replacement of the central point is enhanced, then save the better central point and does the same replacement operation for the next element in center point. After the replacing operation for all route groups, we update the best center point. Later, continue the replacement cycle until the quality of clustering not change. However, there is a problem: in a standard PAM, clustering center is easy to fall into local optimal center combination, which can lead to an unreasonable division of sub-problem reducing the ability of the CC framework and directly affects the quality of the final solution. For a briefly exposition, we will illustrate in the following
