Solar Irradiance Prediction Based on LSTM-CNN Based Hybrid Approach

Integration of solar power sources into existing power grids is hindered by their inherent variability. Solar irradiance forecasting is therefore crucial for optimizing energy generation, reducing costs, and enhancing grid stability. This study explores various hybrid deep learning models, specifically LSTM-CNN and CNN-LSTM, and compare their performance with LSTM and CNN models at multiple time horizons and multiple locations in India. The models utilize a diverse set of common meteorological features. Results demonstrate that hybrid models outperform standalone models, particularly for hourly forecasting, leveraging both temporal and spatial dependencies in the data. Additionally, while standalone LSTM models perform well with univariate inputs for hourly forecasts, hybrid models exhibit superior predictive capabilities when trained on multivariate datasets. The methodology and results provide improvement in renewable energy forecasting, with implications for enhanced grid management and energy market participation. Index Terms—Energy prediction, Solar, LSTM, CNN