Future-ready power grids: From variability to predictability with scalable AI for PV energy integration

The increasing integration of renewable energy, particularly solar power, presents challenges in maintaining grid stability due to fluctuations in power generation and voltage variations. This issue is especially important because solar energy is highly variable and depends on weather conditions, which makes it difficult to keep the power grid stable and reliable. Therefore, there is a strong need for accurate tools that can help predict these changes and improve the way the grid is managed. This study addresses these challenges by leveraging AI-based methods that combine climate data analysis and power grid simulations. The analysis relies on key environmental variables: solar irradiation, temperature, and wind speed to predict two critical outputs: power and voltage levels across the network. Among the tested models, Support Vector Regression (SVR) gave the best performance for power prediction. On the IEEE 123-bus Network, SVR achieved an RMSE of 183.07 and an MAE of 169.15, remaining well within the acceptable margin of 400 kW. For voltage prediction, the Long Short-Term Memory (LSTM) model performed best by capturing long-term time dependencies. On the IEEE 123-bus Network, LSTM achieved an RMSE of 0.0133 and an MAE of 0.0104 for Bus 64, staying well below the acceptable error threshold of 0.015 pu. Accordingly, through addressing a real-world challenge in electrical network operation, this study helps energy systems become more flexible and efficient. The proposed approach supports the transition toward a more stable, clean, and intelligent energy infrastructure.