This paper proposes a grid portfolio forecasting model established by an improved genetic algorithm and comprehensively using the grey prediction GM (1, N), BP neural network and multiple regression models. The traditional genetic algorithm is used to optimize the combined prediction constraints and improve the crossover operator and mutation operator in the genetic algorithm, so that the algorithm has stronger global search and convergence ability. The prediction results of a regional grid investment using the combined forecasting model proposed in this paper show that compared with the single forecasting model and the other two combined forecasting models, the proposed model can make full use of the original data and has higher prediction accuracy. This work is supported by National High-tech R & D Program of China (863 Program) (No. 2015AA050101).