The paper aims to study the luggage stacking algorithm to solve the labor-intensive and inefficient problems in the current flight luggage stacking process for check-in. In this paper, a mathematical model of flight luggage with weight, volume and cargo compartment constraints was built. By using clustering, ranking and key point construction strategy, a new algorithm combining K-means clustering and "key point" idea was designed. Experiments on 100 pieces of real passenger baggage showed that the layout plan given by the algorithm was reasonable. The difference of mass between the left and right sides of the stack was less than 1%, which satisfied the space constraint and load balancing constraint of the cargo compartment. The algorithm has the ability to obtain excellent layout scheme in complex environment. The introduction of K-means clustering algorithm establishes a bridge for machine learning algorithm and traditional packing problems, and provides a new idea for design of bin-packing problem algorithm in the future.