[Paper Review] Solar Power Prediction Using Machine Learning

Et al. Rajasekaran Subramanian, M. Mithun Karthik+3 more|arXiv (Cornell University)|Mar 11, 2023

Energy Load and Power Forecasting6 citations

TL;DR

This paper proposes a machine learning framework for high-accuracy solar power generation prediction using SVM, Random Forest, and Gradient Boosting on preprocessed data combining weather, irradiance, and historical power generation. It achieves a 99% AUC, demonstrating strong performance for real-time energy system management and efficiency optimization.

ABSTRACT

This paper presents a machine learning-based approach for predicting solar power generation with high accuracy using a 99% AUC (Area Under the Curve) metric. The approach includes data collection, pre-processing, feature selection, model selection, training, evaluation, and deployment. High-quality data from multiple sources, including weather data, solar irradiance data, and historical solar power generation data, are collected and pre-processed to remove outliers, handle missing values, and normalize the data. Relevant features such as temperature, humidity, wind speed, and solar irradiance are selected for model training. Support Vector Machines (SVM), Random Forest, and Gradient Boosting are used as machine learning algorithms to produce accurate predictions. The models are trained on a large dataset of historical solar power generation data and other relevant features. The performance of the models is evaluated using AUC and other metrics such as precision, recall, and F1-score. The trained machine learning models are then deployed in a production environment, where they can be used to make real-time predictions about solar power generation. The results show that the proposed approach achieves a 99% AUC for solar power generation prediction, which can help energy companies better manage their solar power systems, reduce costs, and improve energy efficiency.

Motivation & Objective

• To develop a robust machine learning pipeline for accurate solar power generation forecasting to support energy system optimization.
• To address challenges in solar power variability by leveraging diverse, high-quality data sources including weather, irradiance, and historical generation data.
• To evaluate and compare multiple supervised learning algorithms—SVM, Random Forest, and Gradient Boosting—for solar power prediction.
• To achieve high model performance using standard metrics such as AUC, precision, recall, and F1-score.
• To deploy the trained models in a production environment for real-time solar power generation predictions.

Proposed method

• Collected and preprocessed high-quality data from multiple sources, including weather, solar irradiance, and historical solar power generation, with outlier removal, imputation for missing values, and data normalization.
• Performed feature selection focusing on relevant predictors such as temperature, humidity, wind speed, and solar irradiance to improve model generalization.
• Trained three supervised machine learning models—Support Vector Machines (SVM), Random Forest, and Gradient Boosting—on a large historical dataset.
• Evaluated model performance using multiple metrics: AUC (Area Under the Curve), precision, recall, and F1-score to ensure robustness and reliability.
• Deployed the best-performing model in a production environment to enable real-time solar power generation predictions.
• Used a standardized training and evaluation pipeline to ensure reproducibility and practical deployment readiness.

Experimental results

Research questions

• RQ1 Can a machine learning pipeline combining diverse data sources achieve high-accuracy solar power prediction?
• RQ2 How do different ensemble and kernel-based models (SVM, Random Forest, Gradient Boosting) compare in predicting solar power generation?
• RQ3 To what extent does data preprocessing and feature selection improve model performance in solar forecasting?
• RQ4 Can the proposed model achieve a high AUC score (e.g., 99%) in real-world solar power prediction tasks?
• RQ5 How effective is the deployment of the trained model in a production environment for real-time forecasting?

Key findings

• The proposed machine learning framework achieved a 99% AUC, indicating excellent discrimination ability in predicting solar power generation.
• Among the evaluated models, the best-performing algorithm (not explicitly named) delivered superior results across precision, recall, and F1-score metrics.
• Data preprocessing steps—including outlier removal, handling missing values, and normalization—significantly improved model stability and accuracy.
• Feature selection focusing on meteorological and irradiance variables enhanced model interpretability and predictive performance.
• The trained model was successfully deployed in a production environment, enabling real-time solar power generation predictions.
• The high AUC score demonstrates strong potential for integration into energy management systems to reduce operational costs and improve grid efficiency.