Abstract: During the training process of Gaussian process,the commonly used conjugate gradient method is difficult to handle the hyper-parameter of the high-dimensional covariance matrix.Aiming at this problem,this paper introduced the teaching-learning-based optimization (TLBO) algorithm to accelerate the training process of Gaussian process.However,the basic TLBO has local convergence phenomena in certain conditions.This paper proposed an improved TLBO (ITLBO) algorithm by modifying the learner phase and appending an outside-reading phase to increase the population diversity so as to improve the global searching ability.Finally this method is applied to measure the methanol conversion rate and the results indicate that the proposed method has a good result and a certain value.