Improved Solar Power Prediction Using CNN-LSTM Models for Optimized Smart Grid Performance

Abstract

During the fourth energy revolution, the integration of Artificial Intelligence (AI) across various technological fields is critical to meet rising energy demands and address the depletion of fossil fuel reserves, leading to the adoption of smart grids. This study aims to enhance power generation capacity and minimize losses in smart grids by accurately predicting parameters. Traditional power grid stations transitioning to smart grids require precise parameter predictions. To achieve this, we employed AI-based machine learning models, specifically Random Forest (RF) and Long Short-Term Memory (LSTM), to predict the parameters of a solar power plant. After initial analysis through graphical visualization, we further refined the LSTM model using an advanced technique: Convolutional Neural Network (CNN-LSTM). Comparative results indicate that the CNN-LSTM model outperforms both the LSTM and RF models. For daily power generation, the CNN-LSTM achieved the lowest Mean Absolute Error (MAE) of 0.1335 and Mean Squared Error (MSE) of 0.0497. Consequently, the application of AI in this study significantly improves the accuracy of parameter prediction, enhancing the performance of basic machine learning models. This advancement supports the development of a robust and efficient power system that reduces power losses and boosts production capacity within the framework of smart grids.