Abstract

Accurate solar energy forecasting is crucial for grid stability and renewable energy integration, but missing data significantly impacts forecasting accuracy. Traditional statistical imputation techniques often fail to capture the complex, nonlinear nature of solar energy variations, necessitating advanced machine learning approaches. This study addresses this critical challenge by proposing a Multi-Input Fuzzy Rules Emulated Network (MiFREN) model for efficient and accurate reconstruction of missing solar power data. Beyond mere imputation, MiFREN leverages expert knowledge through its interpretable fuzzy rule structure, enabling more robust and physically meaningful forecasts. MiFREN demonstrably outperforms benchmark AI models (KNN, Random Forest, LSTM), achieving the lowest prediction error (NMAPE 1.17 percent, NRMSE 4.13 percent). Crucially, this superior performance is achieved with remarkable efficiency: MiFREN's structured rule base significantly reduces model complexity, requiring only 27 weight parameters - a fraction of the parameters needed by comparable models. This inherent efficiency makes MiFREN highly practical for real-time applications. These findings highlight MiFREN's unique capabilities for enhancing solar energy integration, improving grid reliability, and providing valuable insights into solar power dynamics through its transparent and readily understandable rule base.