Research On Nearly Orthogonal Latin Hypercube Experimental Designs For Multi-Agent

Computer simulation and experimentation,which is a high-quick and efficient Experime-ntal method,has widely been used in military,aerospace engineering,bioengineering and medicine in recent years.In these fields,a common characteristic of the above-metioned areas is the numerous factors—many of which are uncertain and complex correlative between different factors.Conducting a comprehensive experimental design on these vast variables is prohibitive regardless of manpower,financial and material capabilities.Thus,often a small subset of the factors is chosen for experimentation.The experimental designs developed in this dissertation are based on uniform design,orthogonal Latin hypercube design and nearly orthogonal Latin hypercube design,and this dissertation provides a widely universal experimental design methods.The contents of this dissertation are as follows:(1)Discussing different methods of experimental designProviding disadvantages and advantages about completely random design,orthogonal design,uniform design,orthogonal Latin hypercube design and nearly orthogonal Latin hyper-cube design.Based on these theories,the new or improved algorithm about experimental design is proposed later.(2)A new method is proposed to improve the NOLH algorithm in this dissertationUD and OLHC has been proved to be high-quick and efficient experimental design meth-od.However,UD only considers "space-filling" and OLHC just takes in account of Orthogo-nality,unfortunately,these two algorithms don't meet some requirements.On this basis,the NOLH algorithm,which fully considers the "space-filling" and orthogonality,is proposed by Cioppa.Based on the Cioppa's algorithm,a new design containing discrete variables is proposed in this dissertation.The Kenny Q.Ye's method to construct OLHC matrix is proposed and improved.(3)Implementation of the NOLH algorithmOn the basis of numerous experiments,and according to the constraints of orthogonality and uniformity criterion,this dissertation contains both the theory underlying these designs and the details necessary to construct them through JAVA programs,and puts forward how to adjust and implement the second experiment.(4)NOLH algorithm validation and applicationFirst of all,the dissertation contains an application of this methodology on a known non-linear response surface.The corresponding regression analysis is done in R by using forward and backward stepwise regression with AIC.A comparison is made between OLHC and NOLH.At the same time,the resonality of the second experiments is proved.Secondly,based on the SEAS platform,this dissertation produces test data though NOLH for experimental scenarios.Then,results are analyzed and compares with the actual value,and further verify the rationality and feasibility of NOLH.