Methane yield prediction of ultrasonic pretreated sewage sludge by means of an artificial neural network

Highlights: • Anaerobic digestion of ultrasonic pretreated sewage sludge was investigated. • Pretreatments were applied at 0.5 W/mL and in the range of 0.5–240 min. • Sonication of more than 30 min did not significantly increase the methane yields. • The cumulative methane yields were predicted via Artificial Neural Network model. In this study, anaerobic digestion of sewage sludge with different ultrasonic pretreatment (USp) conditions was investigated. USp were applied at a constant ultrasonic density of 0.5 W/mL with sonication times of 0–240 min. While the methane yield of the untreated (control) reactor was 170.1 ± 4.7 mL/g volatile solids (VS), the highest methane yield was 266.1 ± 7.5 mL/g VS in the reactor where sonication was applied for 120 min. Actual specific energy input (SE_A)/nominal specific energy input (SE_N) ratios after USps were measured for pretreatment yield. These values varied between 77.0 and 14.87% according to different sonication times. Sonication times of more than 30 min did not significantly increase the methane yield due to possible reflocculation. After the USps, various cumulative methane yields were predicted via the modified Gompertz Model, modified Logistic Model, and Artificial Neural Network (ANN) model (Scenario 1); of these the ANN model made predictions that were closest to the experimental data. In the other part of the study, ANN was trained by experimental pretreatment conditions and methane yields were successfully predicted by taking different USp parameters (different sonication times, SE_A values and incremental soluble chemical oxygen demand % values) as input variables (Scenarios 2, 3 and 4). Regression analyses showed values close to 1, indicating that the prediction of the ANN model correlated linearly with experimental data.