Research On RoboCup Passing Ball Based On Partial Least Square

With the rapid development of computer technology,research on Multi-AgentSystem has gradually became a key research point in the discipline which is related toArtificial Intelligence.RoboCup provides a standard experimental platform for thestudy of Multi-Agent System in decision-making and collaboration,whichcharacterizes competition and cooperation coexist between agents,dynamicenvironment, communication limited and with random noise etc.It is extremely important that agents play the game strategically.Pass is aconcentrated embodiment of mutual cooperation and collaboration amongagents,which must have important internal relationship with the game.The papermainly aimed at the shortcoming of conventional mode of team development,carrieson pass data analysis research by the method of combining qualitative analysis andquantitative analysis with the idea of Data Mining,tries to make a mathematicalmeasurement as a scientific guidance when agents making high level decisionbetween the pass and the game.The paper carries the work in the research pattern of “data,model,conclusion,decision”.The workflow of this paper are as follows:Firstly,analyzes the structure of log file detailedly and sets the distance intervalof every kind of pass use qualitative analysis method.Then defining relevant datastructure based on the purpose of study and the storage rules of log file, analyzing logfile using C language program and achieving batching processing of log file byFindFile algorithm under the windows system.55field simulation game log files areanalyzed as the experimental data in this study.Secondly,observed data has been made correlation analysis by correlationcoefficient method and point out the defect of simple correlation coefficient,expoundthe theory of multiple linear regression in detail,using multiple regression method tofit the regression equation,the significant test of regression equation and regressioncoefficient has been made.The result shows that regression equation pass the test andpartial regression coefficients does not pass the test,found the problem ofMulticollinearity in observed data.Thirdly,detailedly summarize the development history and application situationof Partial Least Squares which taking the thought of extracting component and avoidthe Multicollinearity problem of explanatory variables effectively,so select PLS as thefinal model,PLS and multiple linear regression are compared,result is achievedthrough the auxiliary analysis technique of Partial Least Squares.It is concluded that long pass plays a major role in a game for passing ball bycombining partial theoretical result with the actual game scene.At last,taking the conclusion as the basis of agents’ decision,then doing corresponding modification inthe project code,that is,enlarge the weight of executing long pass.