Efficient Method for Photovoltaic Power Generation Forecasting Based on State Space Modeling and BiTCN.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Meteorological Forecasting

Background:

• Photovoltaic (PV) power generation is vital for global carbon reduction and the new energy sector.
• Accurate PV output forecasting is essential for efficient energy management due to meteorological influences.

Purpose of the Study:

• To develop an optimal hybrid forecasting strategy for PV power generation.
• To enhance prediction accuracy by integrating advanced deep learning and statistical models.
• To improve the correlation coefficient (R²) and reduce forecasting errors.

Main Methods:

• Integration of bidirectional temporal convolutional networks (BiTCN), dynamic convolution (DC), and bidirectional long short-term memory networks (BiLSTM).
• Development of a novel mixed-state space model (Mixed-SSM) combining SSM, MLP, and MHSA for temporal, nonlinear, and long-term feature capture.
• Application of Pearson and Spearman correlation analyses for feature selection and K-Means++ for data enhancement.

Main Results:

• Feature selection improved the prediction correlation coefficient (R²) by at least 0.87%.
• K-Means++ enhanced input data features, achieving a maximum R² of 86.9% and a 6.62% R² gain.
• The proposed hybrid model achieved a mean absolute error (MAE) of 1.1%, root mean squared error (RMSE) of 1.2%, and an R² of 89.1%.

Conclusions:

• The proposed hybrid forecasting strategy demonstrates superior performance compared to existing BiTCN variants.
• The model effectively forecasts PV power output, contributing to efficient energy management.
• This approach supports low-carbon, safe grid operation through accurate renewable energy predictions.