Catalytic Gasification of Biomass usingArtificial Neural Network (ANN) Technique

Abstract

Energy demands of the growing population coupled with environmental drawbacks of using fossil fuels and its ever-depleting reserves has given primary importance to alternate energy sources such as Biofuels leading to deep research and policy shift in last decade. For the fulfillment of energy demand, due to the arrival of fossil fuels, the world has been slowly evolving towards the modern era. Bio-fuels production through catalytic gasification of biomass is an exciting and promising option. Catalytic biomass gasification is a process that involves the alteration of solid biomass constituents into a gaseous fuel, typically a mixture of CO, H2, CO2, and CH4 through high-temperature thermochemical reactions in the presence of catalyst. ANN is the most appropriate modelling technique in which different process, parameters and output could be optimized. More specifically, a methodology of ANN (Artificial Neural Network) will be accomplished to describe the gasification of biomass. Consequently, in this paper, an artificial neural network model was utilized for simulating the reaction mechanism to describe the function of catalytic gasification of biomass. The dataset was collected during a research study on the performance of a gasification system with 315 fuel samples including biomass, coal, and coal–biomass blends using the Levenberg Marquardt (LM) back -propagation and Bayesian Regularization (BR) training algorithms in the Artificial Neural Networks (ANN) domain. The ANN model used 315 experimental samples of fuels data (C, H, N, S, O, MC, Ash, T, VM, LHV and ER was the main concern in the subject). Additionally, MATLAB 2020a is used and Neural Network toolbox. Artificial Neural Network (ANN) 11 input features / 5 output features having multiple neurons in the hidden layers, a denser to predict catalytic gasification of biomass. Output is compared using mean square error and Regression analysis is applied. For the four gas compositions products (CO, CO2, H2, and CH4), sensitivity analysis on the prediction was performed using MIMO network layer. With a hidden layer neuron number choice of between 5 to 50, it reduced the MSE and give regression (R2) 85 to 95 % which show model accuracy. Here you would get a detailed article on how LM algorithm is superior to BR algorithm. The results showed that the L -M algorithm is better than the BR algorithm. The ANN model provided results very close to the experimental data.