Novel dynamic temporal interaction and feature synthesis framework for enhanced solar power forecasting

• Dynamic Temporal Interaction Modeling. • Feature Synthesis Module and Adaptive Learning Framework. Enhanced Forecasting Accuracy and Interpretable Insights. Solar power forecasting is essential to optimize energy production and maintain a stable power grid operation. However, traditional forecasting methods often struggle either due to ineffective data preprocessing or due to a poor feature extraction and selection process, resulting in reduced accuracy. To address the mentioned issues, this research work introduces a novel Dynamic Temporal Interaction and Feature Synthesis (DTIFS) solar power-forecasting framework. The proposed framework utilizes advanced feature engineering techniques such as interaction terms, polynomial transformations, lagged features, and categorical binning to improve prediction accuracy significantly. A detailed exploratory data analysis (EDA) is conducted, and an intelligent feature engineering process is carried out on the acquired dataset, which led to a significant improvement in the model’s prediction accuracy. To assess the performance of the developed framework, several Machine Learning (ML) and Deep learning (DL) models are applied and tested based on several well-known performance evaluation metrics such as mean absolute error (MAE), root mean square error (RMSE) and R². It was found that the Random Forest (RF) model had an MAE of 15.32, an RMSE of 17.90, and an R² of 0.95 without the proposed framework. However, after applying the novel DTIFS framework, the Multilayer Perceptron (MLP) model outperformed, reaching an MAE of 9.281, RMSE of 12.453, and an R² of 0.98, thus outperforming its competing models under identical operating conditions. This study highlights the crucial role of advanced data transformations in enhancing solar power forecasting models, improving accuracy, and facilitating the integration of renewable energy into the grid. The DTIFS framework demonstrates its effectiveness compared to other advanced models, such as RNN, LSTM, and GAN, positioning it as a promising tool for future solar energy forecasting applications.