To enhance the forecasting accuracy and robustness of photovoltaic (PV) power generation, this study proposes a fluctuating shape-adapted seasonal grey prediction model that effectively captures seasonal variation and time-varying amplitude in PV data.
First, a fluctuating shape-adapted seasonal term is designed in the SGMFSA(1,1) model to adjust periodic parameters. Then, the WOA algorithm is utilized to solve for nonlinear parameters, facilitating the accurate identification of optimal estimates within the proposed model. Moreover, the SGMFSA(1,1) model's performance is evaluated on two real-world PV datasets (quarterly and monthly), benchmarked against two statistical models, four machine learning algorithms, and five grey models.
Empirical results show that the SGMFSA(1,1) model consistently outperforms all benchmark models in terms of forecasting accuracy and stability. In Cases 1 and 2, SGMFSA(1,1) achieves average MAPE of 1.8 and 4.2%, reducing errors by 82.5 and 52.8% compared to benchmark models, with APE ranges narrowed by 88.7 and 65.9%, respectively, demonstrating superior accuracy and stability. Probability density analysis further confirms the model's superior robustness.
The proposed model provides an effective forecasting framework for PV power generation, which is critical for improving grid stability, optimizing energy scheduling, and supporting policy and investment decisions in the renewable energy sector.
Considering the limitations of existing forecasting models in handling PV data with volatility and time-varying amplitudes, we design a novel fluctuating shape-adapted seasonal grey prediction model (SGMFSA(1,1)) to address the time-varying volatility of photovoltaic power generation data.
