Military expenditure is a kind of nonlinear time series with macroeconomic characteristics under the complex economic system. Under the background of the study of military expenditure multi-objective combination forecasting, a nonlinear grey neural network metering combined forecasting model based on the elite genetic algorithm (EGA) is proposed. The overall modeling ideas and nonlinear grey neural network operator sub-system and EGA sub-system design methods are given. The NP-complete problem of multi-objective optimization is solved, and the forecasting effect of the model is analysed. Collecting the United States 27 (1990-2016) years of military expenditure data for empirical testing, the result shows that the nonlinear gray neural network operator can improve the accuracy of the model effectively. The EGA is superior to the SGA in the convergence rate and convergence accuracy. The forecasting model which is used to forecast the military expenditure can get higher accuracy and better results.