Research And Implementation Of Airport Gate Intelligent Assignment Method

With the development of civil aviation,civil aviation operation management has brought great challenges.In Airport operation,gate is the key resource,and its assignment results affect the efficiency of airport operation.Therefore,Airport Gate Assignment Problem(AGAP)has always been a research hotspot.In the existing research,the method of pre-assignment of gate has received extensive attention.However,with the increasing number of flights in recent years,the method of pre-assignment of gate is difficult to adapt to the uncertainty of flight arrival.In this paper we proposed a dynamic gate assignment method.The dynamic AGAP is modeled as Markov Decision Process(MDP),and the policy function is modeled as deep neural network.The deep reinforcement learning(DRL)method is used to train the policy network.A demonstration system of DRL based AGAP is designed and implemented.The specific work is as follows:(1)DRL is still in starting to solve AGAP.Firstly,the simplified model of AGAP is studied in this paper,exploring the feasibility of DRL based AGAP.The classical optimization objective and constraints of AGAP in ideal airport environment are used to model the optimization problem and transforming optimization problem into MDP.The training method DRL-AGAP strategy network based on policy gradient algorithm is studied.Comparing the training results with the optimization software Gurobi,the results show that the proposed method has a significant improvement in the calculation speed when the effect is nearly equal to Gurobi.The feasibility of combining DRL with the AGAP is verified,and the method is applicable to the dynamic AGAP.(2)Based on AGAP,considering the actual rules of the airport,the research of DRL based algorithm is studied.Compared with the simplified model,the study of AGAP considering airport actual rules is extended in the following aspects.First,in the optimization modeling of the AGAP,the optimization objective considers the fixed gate rate with the weight of the number of passengers,and the constraints are modeled based on the actual rules of the airport.Second,in the study of the transformation of optimization problems into Markov decision processes,the modeling of state space is more complicated due to the adjustment of optimization objective and the increasing of constraints.Thirdly,Convolutional Neural Network(CNN)is selected as the policy network.We use the actual data of an airport to train policy network,and compare the training results with the traditional algorithms.The proposed algorithm has certain advantages in terms of effect and calculation speed.(3)Based on the algorithm research,a demonstration system of DRL based AGAP is designed and implemented.We analyze the design goals and various needs.On this basis,the system is divided into three layers: Data access layer is used to store Airport data;business logic layer is used to make airport gate assignment decision;presentation layer is used to display the results of gate assignment to users intuitively.We design and explain the implementation of each level.The system can visually represent the results of gate assignment,and achieve the effect of demonstrating the characteristics of the algorithm.