Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Solar Radiation Prediction in the UTEQ based on Machine Learning Models

Jordy Anchundia Troncoso, Ángel Torres Quijije, Byron Oviedo, Cristian Zambrano-Vega

TL;DR

The study addresses forecasting solar radiation at UTEQ by exploiting 5-minute pyranometer data collected since 2020. It systematically compares linear, kernel-based, and ensemble ML models, using correlation-driven feature selection (notably Temperature) and standard error metrics ($MSE$, $RMSE$, $MAE$, $R^2$) for evaluation. Ensemble methods, especially Gradient Boosting Regressor and Random Forest Regressor, provide the most accurate forecasts by capturing nonlinear radiation patterns, with $R^2$ approaching $0.72$–$0.74$ and low $MSE$ values. A Streamlit-based web tool integrates these models with Meteosource temperature forecasts to deliver real-time solar radiation forecasting for practical solar energy management at UTEQ, with potential for generalization to other sites.

Abstract

This research explores the effectiveness of various Machine Learning (ML) models used to predicting solar radiation at the Central Campus of the State Technical University of Quevedo (UTEQ). The data was obtained from a pyranometer, strategically located in a high area of the campus. This instrument continuously recorded solar irradiance data since 2020, offering a comprehensive dataset encompassing various weather conditions and temporal variations. After a correlation analysis, temperature and the time of day were identified as the relevant meteorological variables that influenced the solar irradiance. Different machine learning algorithms such as Linear Regression, K-Nearest Neighbors, Decision Tree, and Gradient Boosting were compared using the evaluation metrics Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the Coefficient of Determination ($R^2$). The study revealed that Gradient Boosting Regressor exhibited superior performance, closely followed by the Random Forest Regressor. These models effectively captured the non-linear patterns in solar radiation, as evidenced by their low MSE and high $R^2$ values. With the aim of assess the performance of our ML models, we developed a web-based tool for the Solar Radiation Forecasting in the UTEQ available at http://https://solarradiationforecastinguteq.streamlit.app/. The results obtained demonstrate the effectiveness of our ML models in solar radiation prediction and contribute a practical utility in real-time solar radiation forecasting, aiding in efficient solar energy management.

Solar Radiation Prediction in the UTEQ based on Machine Learning Models

TL;DR

The study addresses forecasting solar radiation at UTEQ by exploiting 5-minute pyranometer data collected since 2020. It systematically compares linear, kernel-based, and ensemble ML models, using correlation-driven feature selection (notably Temperature) and standard error metrics (, , , ) for evaluation. Ensemble methods, especially Gradient Boosting Regressor and Random Forest Regressor, provide the most accurate forecasts by capturing nonlinear radiation patterns, with approaching – and low values. A Streamlit-based web tool integrates these models with Meteosource temperature forecasts to deliver real-time solar radiation forecasting for practical solar energy management at UTEQ, with potential for generalization to other sites.

Abstract

This research explores the effectiveness of various Machine Learning (ML) models used to predicting solar radiation at the Central Campus of the State Technical University of Quevedo (UTEQ). The data was obtained from a pyranometer, strategically located in a high area of the campus. This instrument continuously recorded solar irradiance data since 2020, offering a comprehensive dataset encompassing various weather conditions and temporal variations. After a correlation analysis, temperature and the time of day were identified as the relevant meteorological variables that influenced the solar irradiance. Different machine learning algorithms such as Linear Regression, K-Nearest Neighbors, Decision Tree, and Gradient Boosting were compared using the evaluation metrics Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the Coefficient of Determination (). The study revealed that Gradient Boosting Regressor exhibited superior performance, closely followed by the Random Forest Regressor. These models effectively captured the non-linear patterns in solar radiation, as evidenced by their low MSE and high values. With the aim of assess the performance of our ML models, we developed a web-based tool for the Solar Radiation Forecasting in the UTEQ available at http://https://solarradiationforecastinguteq.streamlit.app/. The results obtained demonstrate the effectiveness of our ML models in solar radiation prediction and contribute a practical utility in real-time solar radiation forecasting, aiding in efficient solar energy management.
Paper Structure (14 sections, 8 figures, 2 tables)

This paper contains 14 sections, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Data Collection
  4. Data Description
  5. Data Visualization
  6. Variables Selection
  7. Machine Learning Algorithms
  8. Configuration of the Machine Learning Algorithms
  9. Evaluation Metrics
  10. Results
  11. Machine Learning Models Evaluation
  12. Performance of the Solar Radiation Prediction
  13. Interactive tool for Solar Radiation Forecasting
  14. Conclusions and Future Works

Figures (8)

  • Figure 1: Pyranometer installed in one of the Academic Buildings of the UTEQ
  • Figure 2: Actual Solar Radiation on May 2, 2020
  • Figure 3: Actual Solar Radiation on May 2, 2020
  • Figure 4: Heatmap of Correlations between solar irradiance and environmental factors
  • Figure 5: Results of the Machine Learning Models evaluation
  • ...and 3 more figures