Artificial intelligence and machine learning approaches in composting process: A review.

Studies on developing strategies to predict the stability and performance of the composting process have increased in recent years. Machine learning (ML) has focused on process optimization, prediction of missing data, detection of non-conformities, and managing complex variables. This review investigates the perspectives and challenges of ML and its important algorithms such as Artificial Neural Networks (ANNs), Random Forest (RF), Adaptive-network-based fuzzy inference systems (ANFIS), Support Vector Machines (SVMs), and Deep Neural Networks (DNNs) used in the composting process. In addition, the individual shortcomings and inadequacies of the metrics, which were used as error or performance criteria in the studies, were emphasized. Except for a few studies, it was concluded that Artificial Intelligence (AI) algorithms such as Genetic algorithm (GA), Differential Evaluation Algorithm (DEA), and Particle Swarm Optimization (PSO) were not used in the optimization of the model parameters, but in the optimization of the parameters of the ML algorithms.

Modelling and optimization of sewage sludge composting using biomass ash via deep neural network and genetic algorithm.

In this study, the use of Deep Cascade Forward Neural Network (DCFNN) was investigated to model both linear and non-linear chaotic relationships in co-composting of dewatered sewage sludge and biomass fly ash (BFA). Model results were evaluated in comparison with RSM, Feed Forward Neural Network (FFNN) and Feed Back Neural Network (FBNN), and Cascade Forward Neural Network (CFNN). DCFNN produced predictive results with MAPE values less than 1% for all datasets in all experimental designs except one with 1.99%. Furthermore, the decision variables were optimized by Genetic Algorithm (GA). The desirability level obtained from the optimization results was found to be 100% in a few designs and above 95% in all other designs. The results showed that DCFNN is a reliable and consistent tool for modeling composting process parameters, also GA is a satisfactory tool for determining which outputs the input parameters will produce in an experimental setup.

Modeling and optimization of process parameters in co-composting of tea waste and food waste: Radial basis function neural networks and genetic algorithm.

In this study, the effects of co-composting of food waste (FW) and tea waste (TW) on the losses of total nitrogen (TN), total organic carbon (TOC), and moisture content (MC) were investigated. TW and FW were composted separately and compared with the co-composting of FW and TW at different ratios. While the MC losses were close to each other in all processes, the lowest TN and TOC losses were found in the composting process containing 25% TW as 26.80% and 40.11%, respectively. Moreover, Radial Basis Function Neural Networks (RBFNNs) were used to predict the losses of TN, TOC, and MC. The outputs of RBFNN were compared with Response Surface Methodology (RSM), Support Vector Regression (SVR), and Feed Forward Neural Network (FF-NN). In addition, the optimal parameter values were determined by Genetic algorithm (GA). As a result, it will be possible to simulate and improve different co-composting processes with obtained data.

Evaluation of a cascade artificial neural network for modeling and optimization of process parameters in co-composting of cattle manure and municipal solid waste.

The present study was carried out to improve, test, and validate the Cascade Forward Neural Network (CFNN) for co-composting of municipal solid waste (MSW) and cattle manure (CM). Composting was performed in vessel pilot-scale reactors with different CM rates for 105 days. The CFNN used 5 input variables containing CM and MSW mixture combinations, and 1 output for each of the compost quality parameters. The CFNN results were compared with Response Surface Methodology (RSM) and Feed Forward Neural Network (FFNN) results. Multi-objective optimization process using Genetic Algorithm (GA), the total desirability, which has a much better value than the RSM, was obtained as 0.4455 and the CM ratio and processing time were determined as approximately 23.39% and 104.86 days, respectively. It is concluded that CFNN is a unique modeling tool, exhibiting superior modeling and prediction performance in MSW and compost modeling for CM.

Identifying the maturity of co-compost of olive mill waste and natural mineral materials: Modelling via ANN and multi-objective optimization.

In this study, olive mill waste (OMW) and natural mineral amendments were co-composted to evaluate the compost maturity efficiency. The results were modelled by Feed-Forward Neural Networks (FF-NN) and Elman-Recurrent Neural Networks (ER-NN) and compared Response Surface Methodology (RSM). According to RSM produced a prediction error of more than 10% while Neural Networks (NNs) models were <2%. From, multi-objective optimization, the most suitable materials were expanded vermiculite and pumice with overall desirabilities of 0.60 and 0.56, respectively. The optimum amendment ratios were achieved with 14.3% of expanded vermiculite and 16.0% of pumice for OMW composting. Multivariate Analysis of Variance (MANOVA) results indicated that the materials had a strong effect on composting in parallel with the optimization results. NNs were predictors with superior properties to model the composting processes, can be used as modeling tools in many areas that are difficult and costly to perform new experiments.

A multilayer perceptron-based prediction of ammonium adsorption on zeolite from landfill leachate: Batch and column studies.

In this study, multilayer perceptron (MLP) artificial neural network was used to predict the adsorption rate of ammonium on zeolite. pH, inlet ammonium concentration, contact time, temperature, dosage of adsorbent, agitation speed, and particle size in the batch experiments were used as independent variables while flow rate and particle size in column mode were investigated. In MLP application, different architecture structures were tried and the architecture structures with the highest predictive performance were determined. To comparatively evaluate the predictive capabilities of MLP based prediction tool, Response Surface Methodology (RSM) was utilized. When the results were evaluated with Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) values (<1%) for almost all experiments, it was seen that MLP-based prediction tool produces better predictions than RSM. The scatter plots showed that predictions and actual values were quite compatible. Both regression and determination coefficients were interpreted by creating a regression of the predictions against the actual values and these coefficients were obtained as pretty close to 1. The outstanding performance of MLP in out-of-sample data sets without the need for additional experiment demonstrate that MLP can be effectively and reliably used in cases where experimental setups are difficult or costly.