Cloud model is a new mathematical representation of linguistic concepts, which shows potentials for uncertainty mediating between the concept of a fuzzy set and that of a probability distribution. This paper utilizes cloud model theory as an uncertainty analyzing tool for noise-polluted signals, which formulates membership degree functions of residual errors that quantify the difference between the prediction from simulated model and the actual measured time history at each time interval. With membership degree functions a multi-objective optimization strategy is proposed, which minimizes multiple error terms simultaneously. Its non-domination-based convergence provides a stronger constraint that enables robust identification of damages with lower damage negative false. Simulation results of a structural system under noise polluted signals are presented to demonstrate the effectiveness of the proposed method.