Gate Assignment Problem Research Via Deep Reinforcement Learning

In the regular operation of the airport,the stand is one of the most valuable assets.However,the speed of airport construction is far behind the increase number of flights,and the conventional airplane stand scheme is increasingly difficult to meet the needs of the airport.The allocation result of airport stands is directly related to the interests of all parties in the air transport system.A good stand allocation scheme can reduce the daily scheduling cost and aircraft maintenance cost of airport management,reduce the risk of airport,and improve the travel comfort and satisfaction of passengers.Some uneconomical and unreasonable distribution schemes,once put into use,may lead to misunderstanding and even conflict between airports and passengers,airports and airlines.Therefore,it is of great significance to study the airport stand allocation algorithm,and finally to provide airport managers with high-quality stand allocation scheme.The main work of this paper is as follows:Firstly,the operation of Haikou Meilan International Airport is analyzed and studied.The main reasons for the lack of robustness of airport stand allocation are confirmed as follows: first,when allocating stands,there is not enough interval between successive flights of the same stand plus a large number of delayed aircraft.Second,the airport does not consider the arrangement of parking spaces properly.With time flow,the longer time-parking,the higher chance to change the location.That’s why only one out of the thousandth of the planes with downtime less than 4 hours in half a year of 2018 have changed their stands.Secondly,after the analysis of the airport operation,aiming at the present problems,a relatively complete model of GAP(Gate Assignment Problem)is established.Based on the comprehensive consideration of various common single optimization objective evaluation functions,an evaluation function of multi-objective optimization which fits for the actual situation of Haikou Meilan International Airport is proposed.Through the experiment on real data and simulate data,with the same optimization algorithm,the evaluation function has a synchronous improvement in function value and real allocation effect compared with the artificial result.It shows that the evaluation function can fit the actual needs well.Thirdly,after expounding the deep learning and reinforcement learning,this paper proposes an algorithm of GAP based on deep reinforcement learning,and then explains the operation process of the algorithm in detail.The algorithm is compared with a swarm optimization algorithm.Through the simulation of real data,the algorithm is 9.5% ahead of the result of artificial allocation,5.2% ahead of the particle swarm optimization(PSO)as well as Genetic Algorithm(GA)and 3.2% ahead of Tabu Search(TS).