TY  - EJOU
AU  - Ma, Chenglian 
AU  - Han, Rui 
AU  - An, Zhao 
AU  - Hu, Tianyu 
AU  - Jin, Meizhu 

TI  - Weather-Driven Solar Power Forecasting Using D-Informer: Enhancing Predictions with Climate Variables
T2  - Energy Engineering

PY  - 2024
VL  - 121
IS  - 5
SN  - 1546-0118

AB  - Precise forecasting of solar power is crucial for the development of sustainable energy systems. Contemporary forecasting approaches often fail to adequately consider the crucial role of weather factors in photovoltaic (PV) power generation and encounter issues such as gradient explosion or disappearance when dealing with extensive time-series data. To overcome these challenges, this research presents a cutting-edge, multi-stage forecasting method called D-Informer. This method skillfully merges the differential transformation algorithm with the Informer model, leveraging a detailed array of meteorological variables and historical PV power generation records. The D-Informer model exhibits remarkable superiority over competing models across multiple performance metrics, achieving on average a 67.64% reduction in mean squared error (MSE), a 49.58% decrease in mean absolute error (MAE), and a 43.43% reduction in root mean square error (RMSE). Moreover, it attained an R² value as high as 0.9917 during the winter season, highlighting its precision and dependability. This significant advancement can be primarily attributed to the incorporation of a multi-head self-attention mechanism, which greatly enhances the model’s ability to identify complex interactions among diverse input variables, and the inclusion of weather variables, enriching the model’s input data and strengthening its predictive accuracy in time series analysis. Additionally, the experimental results confirm the effectiveness of the proposed approach.
KW  - Power forecasting; deep learning; weather-driven; solar power

DO  - 10.32604/ee.2024.046644