Learning Large Neighborhood Search for Vehicle Routing in Airport Ground Handling

Dispatching vehicle fleets to serve flights is a key task in airport ground handling (AGH). Due the notable growth of flights, it challenging simultaneously schedule multiple types operations (services) for large number where each type operation performed by one specific fleet. To tackle this issue, we first represent scheduling as complex routing problem and formulate mixed integer linear programming (MILP) model. Then given graph representation MILP model, propose learning assisted neighborhood search (LNS) method using data generated based on real scenarios, integrate imitation convolutional network (GCN) learn destroy operator automatically select variables, employ an off-the-shelf solver repair reoptimize selected variables. Experimental results show that proposed allows up 200 with 10 simultaneously, outperforms state-of-the-art methods. Moreover, learned performs consistently accompanying different solvers, generalizes well larger instances, verifying versatility scalability our method.