Predicting steady-state biogas production from waste using advanced machine learning-metaheuristic approaches

This research offers a fast and accurate method for measuring the biogas production rate throughout biogas production. An agricultural biogas plant's measurement of eight process variables served as the source of experimental data used to create the models. Biomass type, reactor/feeding, volatile solids, pH, organic load rate, hydraulic retention time, temperature, and reactor volume were utilized in this context. Artificial neural networks (ANN) were developed to evaluate the biogas production rate. The variable selection was carried out using the cuckoo optimization algorithm (COA), multi-verse optimization algorithm (MVO), leagues championship algorithm (LCA), evaporation-rate water cycle algorithm (ERWCA), stochastic fractal search (SFS), and teaching-learning-based optimization (TLBO). In this study, the model's size decreased, the important process variables were highlighted, and the ANN models' potential was enhanced for prediction. The proposed COA, MVO, LCA, ERWCA, SFS, and TLBO and ensembles are the outcome of using the abovementioned approaches to synthesize the multi-layer perceptron (MLP). To evaluate the effectiveness of the used models, we have developed a scoring system in addition to employing mean absolute error, mean square error, and coefficient of determination as accuracy criteria. Implementing the COA, MVO, LCA, ERWCA, SFS, and TLBO algorithms enhances the accuracy of the MLP. It is found that some of the used hybrid techniques could provide better prediction outputs than traditional MLP rankings. Additional investigation indicated that the ERWCA is better than the three other algorithms. The biogas production rate was estimated with the greatest precision with R2 = 0.9314 and 0.9302, RMSE of 0.1969 and 0.24925, and MAE of 0.1307 and 0.19591.