Modelling of biogas production process with evolutionary artificial neural network and genetic algorithm

• Main Author: Fakharudin, Abdul Sahli
• Format: Thesis
• Language: English
• Published: 2017
• Online Access: http://psasir.upm.edu.my/id/eprint/68747/1/FSKTM%202018%208%20-%20IR.pdf; http://psasir.upm.edu.my/id/eprint/68747/

In recent years, several researchers have actively pursued the application of machine learning to biogas production processes. The application of artificial neural network (ANN) to generate the production model is used to improve the modelling accuracy. The model output optimisation by genetic algorithm (GA) produces higher biogas production compared to the optimisation using statistical methods. This study utilised the evolutionary artificial neural network (EANN) modelling to improve the model accuracy. The EANN modelling was used to represent the biogas production process. One of the issues of ANN implementation is to correctly select the output activation function in achieving higher output. The EANN used a modified activation function to meet the optimisation requirement. To evaluate the EANN model, 19 samples of experimental data from Zainol on the regression modelling of biogas production from banana stem waste were selected. Thirteen samples were used for training (70%) and six samples were used for testing (30%). The second dataset fromMahanty which consisted of 36 samples on the modelling and optimisation of biogas production from industrial sludge were divided into 25 training samples and 11 testing samples. Meanwhile, 34 samples from Tedesco on the optimisation of mechanical pretreatment of Laminariaceae spp. biomass for the production of biogas were divided into 24 training samples and 10 testing samples. The last dataset from the domain expert containing 143 samples were divided into 100 training samples and 43 testing samples. The model performance was evaluated using root mean square error (RMSE) and coefficient of determination (R2) and the maximum output from the optimisation was compared to the mathematical modelling. The experiment was conducted with 50 trial runs on each dataset and EANN method produced better modelling results compared to the mathematical modelling. The model output from the optimisation using GA also produced better results than the mathematical model and able to limit the maximum output of the back-propagation and Levenberg-Marquardt ANN models which used linear function output.