Production and prediction of biomass syngas using deep learning models

Cihan P., Alfarra F., ÖZCAN H. K., CİNER M. N., ÖNGEN A.

International Journal of Hydrogen Energy, cilt.204, 2026 (SCI-Expanded, Scopus) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 204
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1016/j.ijhydene.2025.153232
  • Dergi Adı: International Journal of Hydrogen Energy
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Artic & Antarctic Regions, Chemical Abstracts Core, Chimica, Compendex, Environment Index, INSPEC
  • Anahtar Kelimeler: CNN-LSTM, Data generation, Deep learning, Explainable AI, SHAP, Syngas prediction, Syngas production
  • İstanbul Üniversitesi-Cerrahpaşa Adresli: Evet

Özet

Accurate prediction of carbon monoxide (CO), hydrogen (H2), and methane (CH4) concentrations in syngas using computational and deep learning (DL) approaches is critical for effective monitoring and management of environmental and industrial biomass conversion processes. This study systematically evaluates multiple deep learning architectures for predicting syngas composition from biomass. A laboratory-generated dataset comprising 3748 samples was preprocessed through data transformation and subjected to exploratory analysis. Feature importance was quantified using Shapley Additive Explanations (SHAP) and Mutual Information (MI). The models tested include Deep Neural Networks (DNNs), Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and a hybrid CNN-LSTM architecture. Model performance was assessed using the coefficient of determination (R2), root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE). Among all models, the hybrid CNN-LSTM exhibited superior predictive accuracy and minimal error for CO (R2 = 0.9877), H2 (R2 = 0.9875), and CH4 (R2 = 0.9831). These results suggest that the CNN-LSTM model is well-suited for future studies involving time-dependent variables in syngas-based recovery processes and may serve as a benchmark for analogous environmental applications.