In relevance vector machine prediction models,the selection of kernel functions and the values of kernel parameters have great influence on the prediction performance of the models.Aiming at this issue,a novel relevance vector machine prediction model is constructed on the basis of multiple-attribute Gaussian kernel functions,and a self-optimal kernel parameter-learning method is proposed to optimize the kernel parameters.Then,the constructed model is used to predict a two-dimensional standard function as well as the effluent quality of a wastewater treatment system.Finally,this model is compared with several models using different kernel functions and several models using different parameter optimization methods by simulation experiments.The results indicate that the proposed model is less sensitive to lower dimension data and has better output accuracy and sparsity in dealing with higher dimension data,and that it shows a satisfying performance in predicting the effluent quality of wastewater.