Abstract: The constitutive model of freezing creep is one of the important reference data for shaft building using freezing method. Uniaxial creep tests under different stress levels were conducted for the artificial freezing clay at -5℃, -10℃, -15℃ and -20℃. So the changing rules of creep curves were obtained under the condition of different temperatures and different stress levels. On this basis, traditional Guess-Newton algorithm has been improved by fuzzy random iterative search and the of the fuzzy random Guess-Newton algorithm steps were obtained, then we use the improved algorithm to optimize the parameters of the generalized Kelvin constructive model, obtained the optimized creep model under various temperature and stress. Engineering experiments show that fuzzy random Guess-Newton algorithm can effectively optimize generalized Kelvin constitutive model parameters, thus making the model better fit to creep values of freezing clay at each stage and accurately representing of the creep characteristics of freezing clay. At the same time, the improved algorithm is more efficient and faster convergent than the traditional algorithm.