[求助] 用lsqnonlin拟合已知活度系数模型，残差大，无法获得最优解，请大神看看代码的问题！ 已有2人参与

求助各位大神！！！        小虫妹在拟合已知NRTL模型时（如图片），用matlab中最小二乘法来拟合实验数据，以求得模型中的两个参数△g12,△g21，目标方程设为OF。 编入的代码如下，运行之后，输出如下，感觉残差值比较大，拟合值好像是局部最优解，而不是全局最优； 改变初始值，对参数值的影响也很大， 初始值的设置也是个难题。        请大神看看，问题该如何解决？？？        另求版内大神是否有高版本1stOp能否代跑一下，给出个初值或者最优解，小虫妹不胜感激，可追加金币！！！ function OF=ObjFunBinary(x)   %目标函数 format long T=[278.15,283.15,288.15,293.15,298.15,303.15,308.15,313.15,318.15]; x1=[0.01217,0.01457,0.01704,0.02085,0.02337,0.02707,0.03041,0.03377,0.03766]; x2=1-x1; R=8.3145; H=19.76; Tm=454.8; gamma_exp=exp(H/R*(1/Tm-1./T))./x1; a=x(1);g1=x(2);g2=x(3);%x(1)根据文献一般取0.2、0.3或者0.47 t12=g1./(R*T); t21=g2./(R*T); G12=exp(-a.*t12); G21=exp(-a.*t21); lngama1=x2.^2.*(t21.*G21.^2./(x1+x2.*G21).^2+t12.*G12./(x2+x1.*G12).^2); gamma_cal=exp(lngama1); OF=gamma_cal-gamma_exp; beta0 = [0.31 2000 8000];  %参数初始值 options.Algorithm = 'levenberg-marquardt'; [beta,resnorm,residual,exitflag,output]=lsqnonlin(@ObjFunBinarynew,beta0,[],[],options) 输出的结果是 Local minimum possible. lsqnonlin stopped because the final change in the sum of squares relative to its initial value is less than the default value of the function tolerance. <stopping criteria details> beta =    1.0e+04 *    0.000044925534703   1.158249518250641   1.085523151964263 resnorm =    3.233501964174955 residual =   Columns 1 through 7   -0.835682305165847   0.375853525304237   0.641305366330478   0.765737066136275   0.366439803004162   0.007880991932048  -0.337551041065019   Columns 8 through 9   -0.624403280363090  -0.870680738315109 exitflag =      3 output =        iterations: 37         funcCount: 168          stepsize: 23.651354916984875      cgiterations: []     firstorderopt: 0.093957635054323         algorithm: 'levenberg-marquardt'           message: 'Local minimum possible. lsqnonlin stopped because the final change in the sum of squares relative to its initial value...' 1305562_1333620757_902.jpg

回复此楼