CASIA OpenIR  > 毕业生  > 博士学位论文
智能飞行模拟训练系统研究
其他题名Research on Intelligent Flight Trainer
耿小兵
学位类型工学博士
导师杨一平
2011-11-29
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业模式识别与智能系统
关键词飞行训练 智能教学系统 行为树 概念知识树知识表示体系 贝叶斯网络 反馈 Flight Training Intelligent Tutoring Systems Behavior Tree Conceptual Knowledge Tree Representation Framework Bayesian Network Feedback
摘要伴随着我国低空空域的逐步开放,通用航空产业将迎来迅猛发展,而传统的飞行训练方法高度依赖教练机与飞行教练,难以满足突如其来的巨大需求。传统的飞行训练模拟器能够模拟飞机和飞行环境,可在一定程度上缓解训练装备和训练场地的压力。然而,飞行技能最好的培训方式是一对一个性化训练,因此最大的挑战来自飞行教练的欠缺。此外,飞行模拟器的功能也未能得以充分利用。 如果能为飞行模拟器增加一定的智能,既可使其替代人类教练的部分工作,又能充分发挥飞行模拟器的潜能,从而可以一举两得地应对通用航空驾驶培训的挑战。从这一思路出发,本文的研究主要围绕智能飞行模拟训练的三个基本问题—教什么、教给谁和如何教—展开。论文的主要工作和贡献归纳如下: 1.提出了一种构建飞机驾驶领域模型的知识表示方法 领域模型针对的是用计算机表示教什么的问题。飞行学员的训练从掌握单个机动动作开始,对应的知识以程序性知识为主,而且根据应用需求,相应的表示方式应该自然、易于编辑和更新,传统的产生式方法既不直观又不够灵活,难以达到这些要求。鉴于此,本文采用了直观、复用性好、扩展性强的行为树以表示飞机驾驶领域的程序性知识。针对行为树不便于管理的弱点,本文根据其所表示的机动动作之间的关系,利用概念知识树知识表示体系加以组织,借助后者提供的良好的扩展接口,将两种知识表示方式自然地结合在一起,提出一种新的知识表示方式,用于飞机驾驶领域模型的构建。 2.构建了针对事件的贝叶斯网络用以飞行学员建模 学生模型针对的是用计算机表示教给谁的问题,即准确评估教学对象。常用的贝叶斯网络方法复杂度高,而飞行训练过程中的评估要求实时性强,二者之间的矛盾可能是造成技能培训领域智能化程度低这一现状的原因之一。针对这一矛盾,本文采用分层、原子化、分步更新等措施将贝叶斯网络细化到了飞行学员训练中触发的事件这种粒度上。考虑到实际应用中可能出现的数据缺失,本文利用EM算法学习网络的参数,并利用仿真学员进行实验,结果证明了这一方法的可行性,同时证实了猜测和疏漏对准确率的影响。 3. 设计了追加反馈生成框架用于支架式教学 支架式教学是建构主义学习理论提倡的教学方式之一。实际空中带飞时,来自教练的提示与帮助即是支架的具体体现。在准确评估飞行学员技能水平的基础上,本文通过自动生成针对性强的同步和末端追加反馈的方式为飞机驾驶模拟训练提供支架。通过反馈编辑器编辑鼓励性反馈、预防性反馈和纠错性反馈,利用运行时引擎将反馈的生成融合到飞行模拟器中,二者一并构成了同步追加反馈的生成框架。另外,通过同步追加反馈内容的选择、讲评内容的推荐以及训练内容的调整,保证了追加反馈的针对性。 4.提出了智能飞行模拟训练系统架构并用原型系统加以初步测试 智能模拟飞行训练系统需要飞行模拟器与智能飞行教练的有机结合。本文将飞行模拟器作为交互接口,并与飞机驾驶领域模型、飞行学员模型、支架式教学模型形成了智能飞行模拟训练系统的整体架构。在XX飞机飞行模拟器原型的基础上,借助开发的著作工具增强其教员控制台的...
其他摘要The airplane has been changing the life of human life deeply since its invention. More than 100 years later, airplane has become a necessary means of transport and been applied broadly both in civil and military aviation. The low-altitude air place of china is opening step by step, which will bring rapid development for general aviation. But the traditional flight training depends greatly on training plane and coach, and is not suitable for the forthcoming huge demand. The traditional flight trainer can relieve the stress on equipment and airport in some degree with the simulation of airplane and flight environment. However, the best way of flight training is one-to-one guidance so the biggest challenge is lack of flight tutor. At the same time, the flight trainer is not made full use of. If it had some kind of intelligence, then it would be capable of doing part of job for flight tutor. And it would answer the challenge from both sides. According to this idea, the research aims to three basic problems of flight training - what, whom and how. The main contribution of this dissertation can be summarized as follows. 1. A hybrid knowledge representation for the domain model of flight training was proposed. The domain model aims at what to teach. The training of student pilots starts from simple flight maneuvers which can be represented in procedural knowledge. It requires a knowledge representation of natural, easy-to-edit-and-update, which is out of the reach of production rule. So Behavior Tree (BT), a knowledge representation with these features, is adopted for the procedural knowledge of flight training. But behavior trees are not easy for management. And Conceptual Knowledge Tree (CKT), a powerful knowledge representation, is able to make the compensation ideally for it. Moreover, CKT has extendable interface which enables its natural combination with BT. Thus, a hybrid knowledge representation for the domain model of flight training is formed. 2. A student pilot modeling based on Bayesian network was constructed. The student model aims to whom to teach. The Bayesian network used in common is too complex for flight training which requires real-time evaluation. So some measures are taken to simplify it. And a Bayesian network for every single event which would appear in flight training was constructed. EM algorithm was used to the learning of parameters for the probable lack of data. Experiments with simulated students prove its feasibility. Two...
馆藏号XWLW1696
其他标识符200818014628034
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6401
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
耿小兵. 智能飞行模拟训练系统研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2011.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20081801462803(13094KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[耿小兵]的文章
百度学术
百度学术中相似的文章
[耿小兵]的文章
必应学术
必应学术中相似的文章
[耿小兵]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。