Prediction of Cumulative biogas yield for FFV using Support Vector Regression

The abundant fruit, food and vegetable (FFV) waste produced by markets can be beneficiated to alternative energy called biogas. However, the variability of FFV feedstock, temperature fluctuation, change in pH as a result of microbial activities often lead to process instability which require continuous monitoring and control. Machine learning approach can provides a less expensive route for the evaluation of benefits and associated risks with gasification process since it does not rely on the experimental method of verification and can predict the real-time condition of the Anaerobic digestion (AD). In this study, Support Vector Regression (SVR) is deployed to forecast the cumulative biogas yield for food, fruit and vegetable waste. Moreover, the effect of varying kernel functions (Linear, Gaussian, and Polynomial) were investigated to handle non-linear relationships between the process parameters and the cumulative biogas yield. Seven (7) input variables; organic loading rate, volatile solids, pH, hydraulic retention time (HRT), temperature, retention time, and reaction volume, were considered. At the same time, cumulative biogas production was the output. The data collected were randomized and subsequently divided in a ratio of 7:3 for training and testing, respectively. The SVR models were assessed based on some verified performance metrics. From the investigation, the Gaussian function performed better than Polynomial while Polynomial performed better than linear function. Specifically, root mean squared error (RMSE), mean absolute deviation (MAD), MAPE, R_sq , rMBE and RCoV of 0.0842, 0.0593, 11.3491, 0.9003, 5.9885 and 0.0905 respectively were reported for Gaussian function with model converging at 77 iterations. This study shows that SVR could be a good candidate for cumulative biogas yield prediction, especially when the optimal kernel function is used.