For the identification of nonlinear systems with noise, the traditional T-S fuzzy identification method is difficult to get better results. Therefore the noise signal is regarded as the input of the antecedent together with other input variables of the system. The improved ant lion optimization(ALO) algorithm with dynamic random search and continuous radius contraction is used to optimize the structural parameters of the antecedent. The weighted least square method is utilized to identify the parameters in the consequent. The simulation results show that the proposed method can effectively repress the noise, and achieve better identification effect by using the improved ALO algorithm. Finally, the proposed method is applied to the identification of the thermal model of CZ silicon single crystal growth, and the experimental results show that it is superior to the traditional identification method.