Abstract: In order to overcome some defects of the existing grey multivariate convolution model with new information priority accumulation (GMCN(1,N)), such as the neglect of variable heterogeneity analysis, the weak capability in nonlinear feature extraction, and the mismatch between model parameter estimation and time response function, based on the ideas of variable-order accumulation and discrete grey models and by introducing an additional nonlinear correction term, a variable-order nonlinear discrete grey multivariate model with new information priority accumulation is proposed. Basic properties of the new model are analyzed. Solution structure as well as model parameters are derived. In addition, the quantum particle swarm optimization algorithm is adopted to seek for the optimal accumulation orders. Finally, the proposed model is applied to two practical cases for multidimensional evaluation. The results indicate that the new model outperforms the classic GM(1,N) model, the existing GMCN(1,N) model, and several recently proposed grey multivariate models in terms of both fitting and prediction accuracy, demonstrating better stability and generalization capability.

Key words: