Considering the robust parameter design of multi-response robust parameters in response to the common variation characteristics, an optimization model is constructed in the framework of multi-task Gaussian processes (MTGP) modeling, combining the mass loss function and the response uncertainty optimization function considering the uncertainty of MTGP (UNMTGP). Firstly, a multivariate Gaussian model considering the response to optimization results is constructed based on the fitting test data of the MTGP model. Secondly, the objective function of uncertain optimization considering the output response is proposed to build a multi-response robust optimization model. Finally, the optimal parameter design is obtained by combining the global optimization method.In addition, combining real cases and using the relevant evaluation indexes of the mass loss function, the validity of the method proposed in this paper is demonstrated. The results show that the proposed method can effectively improve the predictive quality of the model and improve the robustness of the output response by taking into account the influence of response covariation and output uncertainty on the optimization results.