1. Journal article: Comparative Evaluation of Statistical, Machine Learning, and Stacking Ensemble Models for Daily Solar Radiation Forecasting 2. DOI: https://doi.org/10.5281/zenodo.21100026 3. Contact information Name: Challa Sai Prakash Institution: VSB-Technical University of Ostrava E-mail: sai.prakash.challa@vsb.cz ORCID: https://orcid.org/0009-0003-8923-2683 Name: Zdenek Machacek Institution: VSB-Technical University of Ostrava E-mail: zdenek.machacek@vsb.cz ORCID: https://orcid.org/0000-0002-6127-0763 4. Dataset archiving (publication) date: 2026-07-01 5. Place of archiving (publication): Ostrava, Czechia 6. Dataset and Code description: original data and code from original research within the project research Platform for Digital Transformation and Society 5.0 and CETPartnership research project funded by the Technology Agency of the Czech Republic under the Sigma Program. Precisely there are 6,7,8,9,10 number of Figures, 4 Datasets of the study: Datasets and Python scripts for for solar radiation forecasting using both statistical and machine learning models. The project compares the performance of multiple forecasting techniques and combines them using a stacking ensemble model for improved prediction accuracy. The objective of this project is to predict daily solar radiation values using different forecasting algorithms and compare their performance. The implemented models include: - SARIMA (Statistical Time Series Model) - LSTM (Long Short-Term Memory Neural Network) - NARX (Nonlinear AutoRegressive with Exogenous Inputs) - Random Forest - XGBoost - Stacking Ensemble Model The project includes separate scripts for training each model, generating predictions, and combining the predictions using ensemble learning. The overall workflow of the project is: 1. Load the training dataset. 2. Perform feature engineering and preprocessing. 3. Train each individual forecasting model. 4. Generate predictions from each trained model. 5. Train the stacking ensemble model using the predictions from the base models. 6. Evaluate the ensemble model on the test dataset. Recommended Python version: Python 3.9+ Main libraries used: - numpy - pandas - scikit-learn - tensorflow - xgboost - statsmodels - matplotlib - joblib Install dependencies using: pip install -r requirements.txt Running the Project - Step 1 Generate engineered features python feature_extensions.py - Step 2 Train individual models python sarimainputstrain.py python lstmtraining.py python narx_training.py python rflearning.py python xgblearning.py - Step 3 Generate predictions python sarimainputtest.py python lstm_inference.py python narx_inference.py python rfinference.py python xgbinference.py - Step 4 Train the stacking ensemble python ensemble_trainingstack.py - Step 5 Run ensemble inference python ensemble_inferencestack.py Outputs of the project produces: - Individual model predictions - Ensemble model predictions - Forecasting results for daily solar radiation - Performance comparison between models Objective - The purpose of this repository is to demonstrate the implementation of multiple forecasting techniques and evaluate how ensemble learning can improve solar radiation prediction accuracy. This package contains Excel data files and Python scripts, where each script processes and data represents computed models and training and testing dataset: Data files: solar_dataset.zip Dataset contains training and testing data of solar radiation parameters ------------------------------------------------------------------------------------------------------------ File Description ------------------------------------------------------------------------------------------------------------ TRAINWIND.csv Training dataset used for model development. TESTWIND.csv Testing dataset used for evaluating model performance. ------------------------------------------------------------------------------------------------------------ solar_radiation_codes.zip Code set contains vatious analysed models of solar radiation simulation and presiction ------------------------------------------------------------------------------------------------------------ File Description ------------------------------------------------------------------------------------------------------------ feature_extensions.py Performs feature engineering and data preprocessing. sarimainputstrain.py Trains the SARIMA forecasting model. sarimainputtest.py Generates predictions using the trained SARIMA model. lstmtraining.py Trains the LSTM neural network model. lstm_inference.py Performs inference using the trained LSTM model. narx_training.py Trains the NARX model. narx_inference.py Generates predictions using the trained NARX model. rflearning.py Trains the Random Forest regression model. rfinference.py Performs prediction using the trained Random Forest model. xgblearning.py Trains the XGBoost regression model. xgbinference.py Performs prediction using the trained XGBoost model. ensemble_trainingstack.py Trains the stacking ensemble model using predictions from the base models. ensemble_inferencestack.py Performs inference using the trained stacking ensemble model. ------------------------------------------------------------------------------------------------------------ 7. Funding: This work was supported by the European Regional Development Fund under the project Research Platform for Digital Transformation and Society 5.0 CZ.02.01.01/00/23_021/0012599 within the Jan Amos Komensky Operational Program supported by the Ministry of Education, Youth and Sports and co-financed by the European Union. This work was supported by the project Enhance Europe – Energy Harvesting Collectors for Urban Road Pavement (Project No. TQ06000003), funded by the Technology Agency of the Czech Republic under the Sigma Program. This research was also carried out within the CETPartnership (Clean Energy Transition Partnership) under the 2023 joint call for research proposals, co-funded by the European Commission (Grant Agreement No. 101069750) and by the funding organizations listed at https://cetpartnership.eu/funding-agencies-and-call-modules.