摘要
<正> (3)计算结果与讨论本文用 Gauss 型峰的合成来模拟各类典型的重叠峰谱图,如图1至图14所示。其中,第 i 成份峰的理论值用(α_i,β_(M+i)~2,β_i)表示,同时它还表示该成份峰对谱图曲线Ⅰ(x)的贡献项α_iexp〔-β_(M+i)~2(x-β)~2〕。这些谱图基本上代表了一般常见的情况,因此可以用以检验 DRAT 计算方法的可行性和收敛性。为了照顾到普适性,Ⅰ(x)和 x 均处理成无因次量。Ⅰ_0为基线常数。①对约束(?)>0作用的验证约束(?)>0对方法收敛性的影响结果见表1。从表中的计算结果的比较,可以看出,约束对方法的收敛性有着较大的影响。若不考虑约束(相当于不作绝对变换 T。
In this paper,the optimizing method concerned with computer resolution of overlap-ping peaks in spectrum analysis is studied.It is found reasonable to treat the optimizingproblem as a constrained one.A DRAT(Dimension Reduction and Absolute value Trans-formation)Method is proposed,which through the establishment of a minimum set ofthe linear parameters in the spectrum model and its absolute value transformation,makesthe 3M+1(M is the number of peaks in the spectrum profile)dimension constrainedoptimizing problem into an equal 2M dimension unconstrained one.Usingmodified powell method,BFGS method and a couble of both,several tipical analogousspectrum profile consisting of up to seven Gaussian bands were resolved succesively on aIBM-PC micro-computer.Both theoreticla analysis and calcualtion results show that treatment of the spectrumprofile fitting as a constrained optimizing problem can efectively avoid the convergencemethod deversing or converging to a“false minimum”due to difficulties of a good initialestimates of the component parameters.The results and comparision demonstrate that theDRAT Method,while on a constrained optimizing scheme and independent of the initialestimates of the M+1 linear parameters in the spectrum model,has a strong convergentquality so it has wild range of the available initial parameter estimates and wild rangefeasibility in computer resolution of overlapping spectrum bands.
出处
《计算机与应用化学》
CAS
CSCD
1989年第1期25-34,共10页
Computers and Applied Chemistry
关键词
谱图解析
重叠峰
计算法
Optimizing method
Computer resolution
Spectrum analysis
Overlaping peaks
