Hydrogen production via biomass gasification, and modeling by supervised machine learning algorithms

Abstract

Prediction of clean hydrogen production via biomass gasification by supervised machine learning algorithms was studied. Lab-scale gasification studies were performed in a steel fixed bed updraft gasifier having a cyclone separator. Pure oxygen, and dried air with varying flow rates (0.05–0.3 L/min) were applied to produce syngas (H2, CH4, CO). Gas compositions were monitored via on-line gas analyzer. Various regression models were created by using different Machine Learning (ML) algorithms which are Linear Regression (LR), K Nearest Neighbors (KNN) Regression, Support Vector Machine Regression (SVMR) and Decision Tree Regression (DTR) algorithms to predict the value of H2 concentration based on the other parameters that are time, temperature, CO, CO2, CH4, O2 and heating value. The highest hydrogen value in syngas was found around 35% vol. after gasification experiments with higher heating value (HHV) of approximately 3400 kcal/m30.05 L/min and 0.015 L/min were the optimum flow rates for dried air and pure oxygen, respectively. In modeling section, it was observed that H2 concentrations were being reflected effectively by the concentrations estimated through the proposed model structures, and by having r2 values of 0.99 which were ascertained between actual and model results.