Evolutionary Real-world Item Stock Allocation for Japanese Electric Commerce

Abstract

This work addresses a real-world item stock allocation using evolutionary optimization for Japanese electric-commerce. We use the actual data of items to be ordered, existing warehouses, and order records from customers. The target area is all over Japan. The task is to find the optimal distribution of one thousand items to eight warehouses. The problem has two objectives: minimizing the total shipping cost and minimizing the average number of stocked warehouses. The problem also has constraints, including the warehouse capacities and the maximum possible number of shipping from each warehouse. Since the commonly used uniform crossover tends to be destructive in this problem, we propose four crossovers for the problem: the item, the warehouse, the item uniform, the warehouse uniform crossovers. Experimental results show that the proposed item crossover is suited to solve this problem, and the obtained item stock allocations can significantly reduce shipping and stocking costs compared with a human-made allocation.