Kurly Optimizes Logistics QPS with Genetic Algorithm... Reduces Unique Product Count by Over 10%

Kurly's Data Platform team introduced a Genetic Algorithm to improve the efficiency of QPS (Quick Picking System), a core process at logistics centers. QPS is an automated process that distributes picked baskets in real time, and processing speed depends on the number of unique products contained in a basket. This is because a basket containing 5 different items such as apples, persimmons, and tangerines takes longer to scan and move than a basket containing 5 apples.

Optimization Goal and Algorithm Application

Previously, the order-product matrix was arranged in a diagonal matrix form to group adjacent orders, but Kurly modeled this as an Open shop scheduling problem and applied a genetic algorithm. While a genetic algorithm does not guarantee an optimal solution, it is effective at quickly exploring better solutions by repeatedly selecting superior solutions and performing crossover and mutation.

In particular, to improve computational efficiency, the starting generation was carefully constructed. Similar orders within the same Batch were first grouped together to raise initial fitness, and then optimization proceeded by allowing only the top 50% with the fewest unique products to survive and generate the next generation.

Results and Future Plans

Based on experiments using past actual order data, applying the Genetic Algorithm (GA) reduced the number of unique products by approximately 10% or more compared to the existing method, regardless of order scale. This leads to improved basket processing speed. This research is scheduled to be presented at IEEM 2022, held in Malaysia at the end of 2022, and Kurly is currently conducting validation using a Digital Twin before applying it in the field.