Solar Power Prediction Using Dual Stream CNN-LSTM Architecture

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Power Grids
• Machine Learning for Energy Forecasting

Background:

• High solar power penetration offers economic and environmental benefits but poses operational challenges due to generation intermittency.
• Accurate solar power generation prediction is crucial for reliable grid operation and high-quality electricity supply.
• Existing methods struggle with the complex spatio-temporal dynamics of solar power generation.

Purpose of the Study:

• To develop an advanced deep learning model for precise short-term solar power prediction.
• To address the challenges of intermittency and randomness in solar power generation.
• To improve the operational efficiency and planning of power systems with high solar energy integration.

Main Methods:

• Introduction of a dual-stream convolutional neural network (CNN) and long short-term memory (LSTM) network with a self-attention mechanism (DSCLANet).
• CNN extracts spatial patterns, while LSTM captures temporal features from solar power data.
• Fusion of spatial and temporal features, followed by self-attention for optimal feature selection and fully connected layers for prediction.

Main Results:

• DSCLANet demonstrated superior performance on the DKASC Alice Spring solar dataset.
• The model achieved significant reductions in prediction errors, with up to 0.0136 MSE, 0.0304 MAE, and 0.0458 RMSE.
• Performance improvements were noted in comparison to recent state-of-the-art methods.

Conclusions:

• DSCLANet effectively enhances short-term solar power forecasting accuracy.
• The proposed deep learning approach offers a robust solution for managing solar energy integration challenges.
• This advancement supports more stable and reliable power systems with increased renewable energy sources.