A hybrid deep learning framework for global irradiance prediction using fuzzy C-Means, CNN-WNN, and Informer models
Artificial intelligence (AI) is revolutionizing solar energy forecasting, enabling precise irradiance prediction for electric solar vehicles (ESVs) to optimize energy efficiency and extend driving range.This study introduces a novel AI-powered hybrid deep learning framework that synergistically comb...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | Cleaner Engineering and Technology |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2666790825001843 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849390682386464768 |
|---|---|
| author | Walid Mchara Lazhar Manai Mohamed Abdellatif Khalfa Monia Raissi Wissem Dimassi Salah Hannachi |
| author_facet | Walid Mchara Lazhar Manai Mohamed Abdellatif Khalfa Monia Raissi Wissem Dimassi Salah Hannachi |
| author_sort | Walid Mchara |
| collection | DOAJ |
| description | Artificial intelligence (AI) is revolutionizing solar energy forecasting, enabling precise irradiance prediction for electric solar vehicles (ESVs) to optimize energy efficiency and extend driving range.This study introduces a novel AI-powered hybrid deep learning framework that synergistically combines fuzzy C-means (FCM) clustering, convolutional neural networks (CNNs), wavelet neural networks (WNNs), and an Informer model to achieve superior accuracy. The FCM layer first groups meteorological data into coherent clusters, reducing noise and isolating localized patterns. CNNs then extract high-level spatial features from each cluster, while WNNs decode multi-resolution irradiance dynamics, capturing both abrupt fluctuations and gradual trends. Finally, the Informer model — equipped with attention mechanisms — identifies long-term temporal dependencies, selecting the most informative timesteps for accurate prediction.The study’s experiments were conducted using a comprehensive dataset sourced from the Photovoltaic Geographical Information System (PVGIS). This dataset spans from January 1, 2005, to December 31, 2020. Data was collected from four climatically distinct cities in the USA: Phoenix, Arizona (desert); Miami, Florida (tropical); Denver, Colorado (semi-arid, high altitude); and Seattle, Washington (oceanic). The proposed CNN-WNN-Informer model achieved average reductions across all cities of 67.7% in t-statistic, 73.9% in Mean Absolute Percentage Error (MAPE), 82.5% in Mean Absolute Bias Error (MABE), and 59.0% in Root Mean Square Error (RMSE), underscoring its significant improvements. By minimizing prediction uncertainty, the framework empowers smarter battery utilization and route planning, bridging the gap between renewable energy and sustainable mobility. This robust performance suggests its potential for integration into intelligent transportation systems and smart grid applications, paving the way for more resilient and energy-efficient urban environments. |
| format | Article |
| id | doaj-art-0fb2e5affad04fe6919daf49a1fb0ccd |
| institution | Kabale University |
| issn | 2666-7908 |
| language | English |
| publishDate | 2025-09-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Cleaner Engineering and Technology |
| spelling | doaj-art-0fb2e5affad04fe6919daf49a1fb0ccd2025-08-20T03:41:26ZengElsevierCleaner Engineering and Technology2666-79082025-09-012810106110.1016/j.clet.2025.101061A hybrid deep learning framework for global irradiance prediction using fuzzy C-Means, CNN-WNN, and Informer modelsWalid Mchara0Lazhar Manai1Mohamed Abdellatif Khalfa2Monia Raissi3Wissem Dimassi4Salah Hannachi5Laboratory of Robotics, Informatics and Complex Systems (RISC), National Engineering School of Tunis (ENIT), University of Tunis El Manar (UTM), B.P No 37, Le Belvedere, 1002, Tunis, TunisiaLaboratory of Robotics, Informatics and Complex Systems (RISC), National Engineering School of Tunis (ENIT), University of Tunis El Manar (UTM), B.P No 37, Le Belvedere, 1002, Tunis, Tunisia; Higher Institute of Information and Communication Technologies (ISTIC), University of Carthage, B.P No 123, Hamam Chatt, 1164, Tunisia; Corresponding author at: Laboratory of Robotics, Informatics and Complex Systems (RISC), National Engineering School of Tunis (ENIT), University of Tunis El Manar (UTM), B.P No 37, Le Belvedere, 1002, Tunis, Tunisia.Higher Institute of Information and Communication Technologies (ISTIC), University of Carthage, B.P No 123, Hamam Chatt, 1164, Tunisia; Laboratory of Advanced Networking and Systems for Energy Research (LaNSER), Research and Technology Centre of Energy (CRTEn), 2050, Hammam Lif, B.P No 95, Borj Cedria, TunisiaDepartment of Mathematics, Faculty of Science, Monastir, Tunis, TunisiaLaboratory of Advanced Networking and Systems for Energy Research (LaNSER), Research and Technology Centre of Energy (CRTEn), 2050, Hammam Lif, B.P No 95, Borj Cedria, TunisiaTeraLab Research in Electronics, and Informatics, Teratech, Tunis, TunisiaArtificial intelligence (AI) is revolutionizing solar energy forecasting, enabling precise irradiance prediction for electric solar vehicles (ESVs) to optimize energy efficiency and extend driving range.This study introduces a novel AI-powered hybrid deep learning framework that synergistically combines fuzzy C-means (FCM) clustering, convolutional neural networks (CNNs), wavelet neural networks (WNNs), and an Informer model to achieve superior accuracy. The FCM layer first groups meteorological data into coherent clusters, reducing noise and isolating localized patterns. CNNs then extract high-level spatial features from each cluster, while WNNs decode multi-resolution irradiance dynamics, capturing both abrupt fluctuations and gradual trends. Finally, the Informer model — equipped with attention mechanisms — identifies long-term temporal dependencies, selecting the most informative timesteps for accurate prediction.The study’s experiments were conducted using a comprehensive dataset sourced from the Photovoltaic Geographical Information System (PVGIS). This dataset spans from January 1, 2005, to December 31, 2020. Data was collected from four climatically distinct cities in the USA: Phoenix, Arizona (desert); Miami, Florida (tropical); Denver, Colorado (semi-arid, high altitude); and Seattle, Washington (oceanic). The proposed CNN-WNN-Informer model achieved average reductions across all cities of 67.7% in t-statistic, 73.9% in Mean Absolute Percentage Error (MAPE), 82.5% in Mean Absolute Bias Error (MABE), and 59.0% in Root Mean Square Error (RMSE), underscoring its significant improvements. By minimizing prediction uncertainty, the framework empowers smarter battery utilization and route planning, bridging the gap between renewable energy and sustainable mobility. This robust performance suggests its potential for integration into intelligent transportation systems and smart grid applications, paving the way for more resilient and energy-efficient urban environments.http://www.sciencedirect.com/science/article/pii/S2666790825001843AttentionCNNDeep learningForecastingInformerLSTM |
| spellingShingle | Walid Mchara Lazhar Manai Mohamed Abdellatif Khalfa Monia Raissi Wissem Dimassi Salah Hannachi A hybrid deep learning framework for global irradiance prediction using fuzzy C-Means, CNN-WNN, and Informer models Cleaner Engineering and Technology Attention CNN Deep learning Forecasting Informer LSTM |
| title | A hybrid deep learning framework for global irradiance prediction using fuzzy C-Means, CNN-WNN, and Informer models |
| title_full | A hybrid deep learning framework for global irradiance prediction using fuzzy C-Means, CNN-WNN, and Informer models |
| title_fullStr | A hybrid deep learning framework for global irradiance prediction using fuzzy C-Means, CNN-WNN, and Informer models |
| title_full_unstemmed | A hybrid deep learning framework for global irradiance prediction using fuzzy C-Means, CNN-WNN, and Informer models |
| title_short | A hybrid deep learning framework for global irradiance prediction using fuzzy C-Means, CNN-WNN, and Informer models |
| title_sort | hybrid deep learning framework for global irradiance prediction using fuzzy c means cnn wnn and informer models |
| topic | Attention CNN Deep learning Forecasting Informer LSTM |
| url | http://www.sciencedirect.com/science/article/pii/S2666790825001843 |
| work_keys_str_mv | AT walidmchara ahybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT lazharmanai ahybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT mohamedabdellatifkhalfa ahybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT moniaraissi ahybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT wissemdimassi ahybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT salahhannachi ahybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT walidmchara hybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT lazharmanai hybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT mohamedabdellatifkhalfa hybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT moniaraissi hybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT wissemdimassi hybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels AT salahhannachi hybriddeeplearningframeworkforglobalirradiancepredictionusingfuzzycmeanscnnwnnandinformermodels |