BACKGROUND AND OBJECTIVES: Leachate recirculation has become a global practice for anaerobic digestion of municipal solid waste. Implementation of artificial neural networks for modeling and prediction of this process still remains challenging. Additionally, there has been a lack of research regarding the generalization capacity of neural networks using the data from other studies. This study aimed to enhance methane production rates and decrease biostabilization time in municipal solid waste treatment. It addressed the research gap in applying and generalizing neural networks to predict biogas production based on laboratory-measured parameters.
METHODS: Two distinct systems were utilized for leachate treatment. System 1 involved collecting the leachate delivered by a new municipal solid waste reactor and transferring it to a recirculation tank. System 2 consisted of passing the fresh municipal solid waste leachate through a degraded municipal solid waste and then returning the obtained liquid back to the waste reactor. The experimental data were employed to develop an artificial neural network to predict methane content and cumulative biogas production. The model was trained and optimized using the experimental data. The effectiveness and generalizability of the optimal neural network were evaluated by using it for the unseen data from other studies, ensuring its ability to make accurate predictions beyond the training dataset.
FINDINGS: The results demonstrated that in System 1, ammonium and chemical oxygen demand concentrations in the leachate progressively increased to high levels. In System 2, the average removal efficiencies for chemical oxygen demand and ammonium were found to be 85 percent and 34 percent respectively. The methane yield in biogas reached 59 liters per kilogram of dry weight, with a corresponding methane fraction of 63 percent. The neural network model showed an excellent performance, with validation performances of 0.716 and 0.634. The overall performance of the dataset resulted in correlation coefficients of 0.9991 and 0.9975. Finally, high correlation coefficients of 0.88 and 0.82 were achieved by incorporating the test data from other studies.
CONCLUSION: Leachate recirculation enhanced the reduction of chemical oxygen demand and the production of methane in bioreactors. Ammonium concentrations initially increased and later decreased due to waste adsorption and bacterial assimilation. The artificial neural network applied for predicting the cumulative methane production from municipal solid waste displayed a robust generalizability when tested on the data from other studies. The neural network was not significantly affected by changes in waste chemical properties, laboratory conditions, and recirculation rate. However, it showed a significant sensitivity to variation of waste mechanical properties.
- Proper selection of input data improves the prediction accuracy of the neural network;
- The mechanical properties of the waste have a significant effect on methane production;
- Understanding methane production requires the study of the combined effects of parameters;
- Generalizability of artificial neural network should be assessed by its testing on different data sets.