Abstract: Current time-varying grey models generally suffer from structural rigidity and insufficient consideration of the intervention effect. To further broaden the application scope of time-varying grey models, this paper proposes a novel fractional time-varying discrete grey model based on the intervention effect (IE-FTVDGM(1, N)). The new model reconstructs the time-varying term by introducing a fractional-order time series to construct a grey model with an adaptive time-varying structure. The reconstructed time-varying term combines integer-order and fractional-order polynomial properties, and can be flexibly adjusted by optimizing a single hyperparameter. In addition, the dual carbon policy is incorporated into the modeling framework as an intervention dummy variable to capture anomalous fluctuations.To determine the optimal hyperparameters, we select the particle swarm algorithm after comparing it with other intelligent algorithms. Finally, Monte Carlo is used to evaluate the robustness of the proposed model. When applied to forecasting China’s new energy vehicle sales, the proposed IE-FTVDGM(1, N) achieves an MAPE of 0.101% and an R2 of 0.9999. It consistently maintains an MAPE below 1% under three different noise levels, demonstrating superior accuracy and stability over all competing models.

Key words: Grey model, Fractional time-varying term, Intervention effect, Prediction of new energy vehicle sales in China