基于XGBoost和时域特征提取的网球比赛实时胜率预测与势头分析  

Live Winning Probability Prediction and Momentum Analysis in Tennis Matches Based on XGBoost and Time-Domain Feature Extraction

作　　者：栗阳 李兆邦 罗廷金 Yang Li;Zhaobang Li;Tingjin Luo(College of Science,National University of Defence Technology,Changsha Hunan)

机构地区：[1]国防科技大学理学院,湖南长沙

出　　处：《建模与仿真》2024年第6期5732-5743,共12页Modeling and Simulation

摘　　要：实时胜率和势头是用于分析网球比赛动态的重要指标,在体育博彩、赛事解说和技术指导中具有应用价值。然而,当前二者的量化方法主要依赖单一模型和静态统计,难以全面捕捉赛事动态。为此,本文引入实时胜率和单次得分价值等概念,基于历史赛事和当前比赛实时数据,采用XGBoost构建分层状态转移模型以预测胜率,量化每个得分点对胜率的影响,在此基础上结合时域特征提取定义比赛双方势头(Momentum),深入分析球员表现与比赛动态变化。Live winning probability(LWP)and momentum are crucial indicators for analyzing tennis dynamics,with applications in sports betting,commentary,and coaching.Current quantification methods,rely-ing on single models and static statistics,fail to fully capture match dynamics.By introducing the concepts of LWP and point impact value(PIV),and using historical and real-time data,we employ XGBoost to build a layered state transition model,predicting win probability and quantifying each point’s impact,based on which time domain feature extraction is used to define player momentum,enabling deeper analysis of performance and match dynamics.

关 键 词：势头 时域特征提取 机器学习 胜率预测 

分 类 号：G84[文化科学—体育训练]