Journal of Heat and Mass Transfer Research

Journal of Heat and Mass Transfer Research

Comparative Analysis of Machine Learning and Deep Learning Stacked Ensembles for Long-term Solar Irradiance Prediction

Articles in Press

Document Type : Full Length Research Article

Authors
Prabhakar Bhandari 1
Rahul Kumar 2
ashwani Kharola 2
Chandra Kishore 2
1 Mechanical Engineering department, K. R. Mangalam University
2 Department of Mechanical Engineering, Graphic Era Deemed to be University, Dehradun, 248002, Uttarakhand, India
10.22075/jhmtr.2026.40901.1936
Abstract
This paper investigates the long-term estimation of All-Sky Surface Shortwave Downward Irradiance (ASD) for the solar-rich urban region of Ahmedabad, Gujarat, India. The study evaluates the predictive performance of three advanced stacked ensemble configurations namely Long Short-Term Memory (LSTM) networks, eXtreme Gradient Boosting (XGB), and Adaptive Boosting (AD) as base learners, paired with a Random Forest (RF) meta-learner. Hyperparameters were systematically optimized via the Optuna framework, while feature interactions were analyzed using the SHapley Additive exPlanations (SHAP) framework. The predictive models were validated using a 20-fold Time-Series Cross-Validation (TSCV) scheme over an 80:20 data split using a suit of error metric. The empirical results reveal that the XGB_RF stacked ensemble model outperforms the competing hybrid frameworks across all statistical metrics. The XGB_RF model achieved an exceptional Coefficient of Determination (R²) of 0.9993, paired with a remarkably low Mean Absolute Error (MAE) of 4.67 W/m² and a Root Mean Square Error (RMSE) of 8.34 W/m². These findings indicate that the proposed XGB_RF framework offers a highly reliable tool for localized solar energy assessment, photovoltaic grid integration, and power management systems.
Keywords
Stacked ensemble
ASD
Forecasting
LSTM
Machine learning
Time-series
Subjects
Journal of Heat and Mass Transfer Research

Articles in Press, Accepted Manuscript
Available Online from 02 November 2026

  • Receive Date 14 April 2026
  • Revise Date 24 October 2026
  • Accept Date 02 November 2026