CASIA OpenIR  > 毕业生  > 硕士学位论文
面向速度滑冰的智能分析及决策技术研究
杨威
Subtype硕士
Thesis Advisor常红星
2022-05-24
Degree Grantor中国科学院自动化研究所
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword速度滑冰 智能分析 深度强化学习 战术决策
Abstract

速度滑冰作为冬奥会最精彩的比赛项目之一,受到了各个国家体育局广泛的关注和重视。利用智能分析及决策技术对速滑项目进行研究,对影响比赛因素的挖掘以及比赛成绩的提升有重要意义,具备很大的实用价值。本文对速度滑冰展开智能分析及决策技术研究,具体研究内容如下:

(1) 针对传统数据分析方法准确性较差和时空分析粒度粗糙等问题,本文提出了一种速滑运动员状态表现综合分析评价方法。该方法通过特征建模构建丰富特征集,然后基于决策树模型设计了重要性分析算法,实现对影响比赛成绩的特征进行重要性分析。最后基于特征重要性设计了综合评价模型,对运动员竞技状态和运动表现的综合评价,提高了评价的准确性。

(2) 针对传统速滑比赛战术决策方法中存在决策规则繁杂、决策效果差以及环境耦合性差等问题,建立了基于深度强化学习的速滑战术决策方法。该方法综合考虑真实场景、物理约束和比赛规则搭建了速滑比赛训练环境,设计滑行奖励、体能奖励以及成绩奖励在内的奖励引导,采用深度强化学习算法进行速滑比赛战术决策,实现了提高竞技水平的高效战术策略学习。

(3) 针对速滑比赛中存在运动员碰撞约束、速度约束以及其他比赛规则等问题,本文提出一种基于知识驱动的深度强化学习算法进行速滑战术决策优化,主要通过基于数据驱动的知识模仿、基于知识的奖励函数设计以及基于知识的动作空间约束实现。该方法基于历史比赛数据,通过监督学习的方式进行知识模仿。根据先验知识和比赛规则进行知识过滤,设计考虑碰撞约束和速度约束在内的奖励函数,提高了战术决策的合理性和有效性。

(4) 根据深度强化学习训练得到的战术策略,对运动员进行战术策略指导与分析。将模型智能生成的策略与具体的圈、段、弯道对应,并与运动员真实比赛战术策略进行时空对应,形成教练员、运动员便于理解和实践的技战术策略指导,完成了模型学习策略的可解释性表达,提高了优化策略对日常训赛的指导作用。

Other Abstract

As one of the most exciting events in the Winter Olympics, speed skating has received extensive attention from various national sports bureaus. The use of intelligent analysis and decision-making technology to research speed skating events is of great significance for assisting the improvement of competition performance. This paper studies the intelligent analysis and decision-making technology for speed skating. The main research contents are as follows:

(1) Aiming at the problems of poor accuracy and rough establishment granularity of traditional data analysis methods, a comprehensive analysis and evaluation method of speed skaters' state performance is proposed. This method builds a rich feature set through feature modeling, and then designs an importance analysis algorithm based on the decision tree model to realize the importance analysis of the features that affect the game performance. Further, a comprehensive evaluation model is designed to evaluate athletes' competitive state and sports performance, which improves the accuracy of evaluation.

( 2) Aiming at the problems of complex decision-making rules, poor decision-making effect and poor environmental coupling in the traditional speed skating tactical decision-making method, a speed skating tactical decision-making method based on deep reinforcement learning is proposed. First, this method builds a speed skating competition training environment considering the real scene, physical constraints and competition rules. Next, we use deep reinforcement learning algorithm to make tactical decision-making in speed skating competitions, and design reward functions such as skating reward, physical fitness reward and performance reward. Then, this method realizes the learning of tactical decision-making strategies that can effectively improve the game performance.

(3) Aiming at the problems of athletes' collision constraints, speed constraints and other competition rules in speed skating competitions, a knowledge-driven deep reinforcement learning algorithm is proposed to optimize speed skating tactical decisions. It is mainly realized through data-driven knowledge imitation, knowledge-based reward function design, and knowledge-based action space constraints. This method performs knowledge imitation through supervised learning based on historical game data. The rationality and effectiveness of tactical decisions can be improved by designing reward functions that take into account collision constraints and speed constraints.

(4) According to the tactical strategy obtained by deep reinforcement learning training, the tactical strategy guidance and analysis of the athletes are carried out. By comparing the tactical strategy learned by the agent with the real strategy of the athlete in time and space, an easy-to-understand tactical strategy guidance can be formed. Through the interpretable expression of tactical strategies generated by the model, the guiding role of the decision-making model for athletes' skating training is improved.

Pages118
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/48512
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
杨威. 面向速度滑冰的智能分析及决策技术研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,2022.
Files in This Item:
File Name/Size DocType Version Access License
杨威-毕业论文.pdf(4891KB)学位论文 限制开放CC BY-NC-SA
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[杨威]'s Articles
Baidu academic
Similar articles in Baidu academic
[杨威]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[杨威]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.