Comparative analysis of AI-based regression models for solar power forecasting: Techniques, preprocessing, and evaluation

Abstract

Accurate prediction of photovoltaic (PV) power output is crucial for improving the efficiency, planning, and stability of solar energy systems. While numerous machine learning (ML) and deep learning (DL) methods have been proposed for this task, practical, implementation-focused resources that guide researchers in model selection, tuning, and evaluation remain limited. This study presents an implementation-oriented comparative analysis of widely used AI-based regression techniques for solar power forecasting. The analysis encompasses key stages including data preprocessing, model selection, performance assessment, and hyperparameter optimization using a 20-year real-world solar dataset with hourly resolution. A set of widely used supervised regression models, including k-Nearest Neighbors, Linear Regression, Ridge and Lasso Regression, Decision Tree, Random Forest, Support Vector Regression (SVR), Gradient Boosting Regressor (GBR), Bagging, Voting Regressor, Multi-Layer Perceptron (MLP)-based Artificial Neural Network, and basic Deep Learning architectures are evaluated in this study. These models are assessed using Python-based tools such as scikit-learn and TensorFlow, with a focus on reproducibility and ease of adaptation for energy system applications. Rather than proposing novel model architectures, the study aims to serve as a practical guide for researchers and engineers, offering insight into the comparative performance and usability of existing regression techniques. The findings emphasize the importance of data quality, preprocessing, and tuning strategies in determining model performance. By combining theoretical clarity, implementation guidance, and empirical evaluation using open-source Python tools, this work provides a reproducible and practical benchmark for AI-driven solar power forecasting, addressing a critical gap in implementation-focused comparative studies.