The work aims to study the goods allocation when the initial storage status of the e-commerce warehouse is not empty, so as to improve the order picking efficiency of the “goods-to-person” picking system. Considering the information of the existing goods on the shelf, the number of vacant goods allocation in the warehouse, the goods to be replenished and the information of the new incoming goods, a decentralized storage strategy based on the degree of goods relevance was proposed to maximize the sum of the relevance of all stored goods on the shelf, thus constructing the mathematical model of dynamic allocation of goods on and off the shelves. Then, a greedy algorithm was designed, and the particle swarm algorithm was improved to optimize the results obtained by the greedy algorithm. The decentralized storage strategy based on the degree of goods relevance could reduce the average number of shelf handling by about 29.32%. Compared with the random allocation strategy, the proposed allocation strategy can effectively improve the picking speed of the entire e-commerce storage center.