Investigations on biomass gasification derived producer gas and algal biodiesel to power a dual-fuel engines: Application of neural networks optimized with Bayesian approach and K-cross fold

Alruqi M., Sharma P., Ağbulut Ü.

Energy, vol.282, 2023 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Volume: 282
  • Publication Date: 2023
  • Doi Number: 10.1016/
  • Journal Name: Energy
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, PASCAL, Aerospace Database, Applied Science & Technology Source, Aquatic Science & Fisheries Abstracts (ASFA), CAB Abstracts, Communication Abstracts, Compendex, Computer & Applied Sciences, Environment Index, INSPEC, Metadex, Pollution Abstracts, Public Affairs Index, Veterinary Science Database, Civil Engineering Abstracts
  • Keywords: Biomass gasification, Emission, Machine learning, Producer gas, Sustainability
  • Yıldız Technical University Affiliated: No


The adoption of sustainable energy sources is a crucial step towards achieving a low-carbon economy and mitigating the effects of climate change. One promising approach is the use of Producer Gas (PG) derived from solid biomass materials, which can be burned as fuel in internal combustion engines to generate power. Biomass gasification, the process of converting solid biomass into PG through thermochemical means, offers a sustainable alternative to traditional fossil fuels. However, using PG in dual-fuel engines poses a significant challenge due to its complex combustion characteristics. Fortunately, modern machine learning techniques offer a promising solution to this problem. In this study, we propose a Bayesian optimized neural network (BONN) to predict the performance and emissions of PG-algal biodiesel (ABD) -powered dual-fuel engines. The BONN is trained using experimental data obtained from a single-cylinder diesel engine retrofitted to run on PG as the primary fuel and algal biodiesel as the pilot fuel. The performance and emissions data are collected under various operating conditions, such as engine load, fuel injection pressure, biodiesel blending ratio, and injection timings. K-cross fold validation was used to reduce the chances of model overfitting while the Bayesian approach was used for hyperparameters finetuning. This strategy helped in the reduction of predicting errors and improved the accuracy of the model. The coefficient of determination was in the range of 0.9421–0.9989 and mean squared errors were in the range of 0.0026–15.77. The mean absolute errors in model-predicted values were in the range of 0.0027–2.945. In all the cases of the prediction results during the model, the test improved upon the model training predictions, indicating a robust generalization and negated the chances of model overfitting. The results demonstrate that the BONN can accurately predict the performance and emissions of the engine, making it a valuable tool for engine optimization and control. This approach offers a promising pathway toward achieving net-zero targets and a sustainable future.