【作者】 李强；

【导师】 杨红雨；

【作者基本信息】 四川大学 ， 工程硕士（专业学位）， 2021， 硕士

【摘要】 航空运输是一个特别容易受气象因素影响的领域,当气象情况较为恶劣,且大面积分布,不满足飞行器正常航行的条件时,现有的空中管制调配手段,主要由管制员进行航向引导,绕飞危险天气,比较严重时,让航班被动地等待,直到计划航线经过的区域恢复容量后才可通行。该情况给人们的出行带来了极大的不便,甚至会造成一定的经济损失。危险天气已经成为航班延误的重要原因。因此,研究危险天气下的空中交通管理策略,具有重要的现实意义。本文对雷暴等危险天气的分布状态进行研究后,总结出了多种分布状态。主要针对散点状分布的危险天气的航行调配问题,提出了改进的强化学习算法,并将改进的算法与启发式算法相结合。首先,针对现有的Q-Learning算法存在的对环境信息部分未知状态的适应能力较差的问题,提出了一种基于条件选择模型的Q-Learning算法。改进后的算法,提高了Agent对环境的适应能力。接下来,将改进后的Q-Learning算法与启发式算法相结合,构建模拟避障实验平台,进而对危险天气下的航班进行改航规划。本文研究主要包括以下内容:(1)分析危险天气下航班改航现状和强化学习对路径规划问题的适用性,尤其在路径避障、航班改航领域。(2)对模拟气象数据进行预处理,提取出危险天气区域,并根据不同的分布特征,将危险天气分为块状分布、带状分布和散点状分布三种分布状态。并提出相应的改航算法和改航策略。(3)针对散点状分布的危险天气情况,提出了改进的Q-Learning算法与启发式算法相结合的算法模型。将算法应用于模拟仿真程序,通过与Dijkstra算法生成的改航路径的仿真数据对比,发现本文方案比使用Dijkstra算法方案的飞行时间、飞行距离分别减少了9.08%、7.32%。综上所述,基于强化学习的算法优化,在危险天气下可获得安全且更加优秀的改航路径,表明了基于强化学习的危险天气改航路径规划具有一定的实时性和科学性。更多还原

【Abstract】 Air transportation is an area that is particularly susceptible to meteorological factors.When the weather conditions are bad and they are also distributed over a large area and cannot meet the normal navigation conditions of the aircraft,the existing air control deployment methods often allow the flight to wait passively until the area where the planned route passes is restored to capacity.This situation brings great inconvenience to people’s travel,and even causes certain economic losses.Dangerous weather has become an important reason for flight delays.Therefore,the study of space traffic management strategies under dangerous weather has important practical significance and practical value.After researching the distribution of dangerous weather such as thunderstorms,this paper summarized a variety of distributions.Mainly aiming at the problem of sailing deployment in scattered points of dangerous weather,an improved reinforcement learning algorithm is proposed,and the improved algorithm is combined with the heuristic algorithm.First of all,in view of the poor adaptability of the existing Q-Learning algorithm to the unknown state of the environmental information,a Q-Learning algorithm based on the conditional selection model is proposed.Although in terms of algorithm efficiency slightly reduced,The improved algorithm improved the Agent’s ability to adapt to the environment.Next,the improved Q-Learning algorithm is combined with the heuristic algorithm to build an obstacle avoidance simulation platform,and then the flight in dangerous weather can be rerouted.The research in this paper mainly includes the following contents:(1)Analyze the status of flight diversion under dangerous weather and the applicability of reinforcement learning to route planning problems,especially in the areas of route obstacle avoidance and flight operations.(2)Preprocess based on simulated weather data to extract dangerous weather areas.According to different distribution characteristics,dangerous weather is divided into block distribution,strip distribution and scattered distribution,and corresponding diversion algorithms and diversion strategies are proposed.(3)Aiming at the dangerous weather situation with scattered points,an algorithm model combining an improved Q-Learning algorithm and heuristic algorithm is proposed.Applying the algorithm to a simulation system,comparing with the simulation data of the aircraft’s randomly generated diversion path,it is found that the optimal solution reduces the flight time and flight distance by 4.66%and 3.66%,respectively,compared with the random solution.To sum up,based on the algorithm optimization under reinforcement learning,a safer diversion path in line with aircraft performance is obtained under dangerous weather,which shows that the dangerous weather diversion path planning based on reinforcement learning has a good experimental effect.更多还原