Abstract
The aim of this research is to investigate the sustainability of waste to energy production using the anaerobic digestion (AD) process. This digestion process has been selected as the ideal waste-to-energy technology owing to minimal effect of carbon dioxide emissions during energy production. Various biomasses such as food waste, sewage sludge, cow and chicken manure were used as feedstock to the AD experiment to produce biomethane. The experimental results of the produced biomethane were used to conduct modelling and simulation using different kinds of computational tools such as Python, R programming, and MATLAB.
An AD experiment was conducted at a laboratory scale to analyze biogas production from diverse substrates such as food waste, cow manure, sewage sludge, and chicken manure, which had an average pH of 7.58. The substrate characterization of the moisture content ranged from 60.32-98.87 %; volatile content was 2.86-24.41%, and total solid was 1.13-39.68%. The experiment was achieved using a biomethane potential (BMP) test for anaerobic digesters containing substrate and inoculum, constantly stirred, and immersed in a water bath at a constant temperature of 370C. AMPTS II was utilised for online measurements of ultra-low flows of biomethane generated by anaerobic digestion. Biomethane data results were collected for up to 29 days.
The experimental data obtained was then modelled using Modified Gompertz, Logistic and Richards models then simulated using AI-Tools: machine learning using Python with Gompertz, Logistics and Richards models, R-programming, and MATLAB. The mathematical modelling and simulation performance was validated by the coefficient of determination (R2) > 0.9; the Modified Logistic model was observed as the best-fitted curve compared to other models. The experimental results obtained from the cow manure substrate illustrated the best fitting curve to the training curve when compared to other substrates. Simulation with machine learning (Python) with all models revealed that the cow manure substrate provided the best fitting curve to the training curve.
Simulation with R-programming was a success as all the graphs plotted were sigmoidal curves, which clearly shows that the model worked well with the biomethane data from all substrates. Simulation-modelling with MATLAB revealed that cow manure substrate produced the highest biomethane compared to other substrates, resulting to the highest average of R2 as 0.997 for training, validation, and test data. Cow manure had the best validation performance at MSE
iii
(mean squared error) of 25.3557 at epoch 1. The AD process using conventional mathematical modelling and simulation/prediction with machine learning using Python language, R-programming, and MATLAB was successfully conducted and implemented as a functional route for predicting and forecasting biomethane production.