To effectively capture the nonlinear and complex patterns inherent in small-sample data, this paper proposed a novel self-adaptive second-order discrete grey model incorporating new information priority accumulation. The modeling mechanism and theoretical properties of the proposed model were systematically examined. To mitigate multicollinearity and enhance the stability of parameter estimation in small-sample contexts, Ridge regression regularization was strategically incorporated into the modeling framework. The computational efficiency of the parameter optimization was demonstrated through comparative benchmarking, which substantiated the superiority of the differential evolution algorithm over conventional methods. Furthermore, the model's robustness and stability were verified via Monte Carlo simulations and sensitivity analyses with varying training set proportions. The proposed model was applied to forecast China's total electricity generation as well as generation from specific energy sources. The results demonstrated that the proposed model achieves higher forecasting accuracy than benchmark models. Finally, the power generation across four modes from 2025 to 2030 was predicted and analyzed.