Artificial neural network (ANN) modeling for the prediction of odor emission rates from landfill working surface. (1st February 2022)

• Record Type: Journal Article
• Authors: Xu, Ankun; Li, Rong; Chang, Huimin; Xu, Yingjie; Li, Xiang; Lin, Guannv; Zhao, Yan
• Is Part Of: Waste management. Volume 138(2022)
• Page Start: 158
• Page End: 171
• Publication Date: 2022-02-01
• Subjects: Landfill -- Odor pollution -- Artificial neural network (ANN) -- Emission rate -- Prediction model; ANN Artificial neural network -- MLPNN Multi-layer perceptron neural network -- COD Chemical oxygen demand -- MSW Municipal solid waste -- ER Emission rates -- NRMSE Normalized root mean squared error -- GA Genetic algorithm -- R2 Coefficient of determination -- L-M Levenberg–Marquardt -- RMSE Root mean squared error -- MISO Multiple input–single output -- VOCs Volatile organic compounds; Hazardous wastes -- Periodicals; Refuse and refuse disposal -- Periodicals; 363.728
• DOI: 10.1016/j.wasman.2021.11.045
• Languages: English
• ISSNs: 0956-053X

Abstract

Graphical abstract: Highlights: Ethanol, methyl sulfide, dimethyl disulfide were identified as typical odor compounds. Established ANN prediction models of landfill odor emission rates with 99 datasets. The performance of ANN models was significantly improved by genetic algorithm. Sensitivity and uncertainty analyses reflect the robustness of ANN models. Abstract: Landfills release significant odorous compounds from the working surface, and their emission rates are crucial for odor and health risk assessment. A total of 99 valid datasets of odor emissions from a landfill working surface were obtained from in situ monitoring for 9 months. Meteorological parameters (temperature, humidity, atmospheric pressure) and waste properties (contents of protein, lipid, carbohydrate, ash, and moisture) were used to construct artificial neural network (ANN) models for the emission rate prediction of typical compounds. The optimal structures and performance of the ANN models were determined by comparing and training with different structural configurations. The ANN models with genetic algorithm (GA) optimization show better performance than those without GA. With the data distribution of input parameters, the ranges of the emission rates of typical compounds were predicted by combining the established ANN models and the Monte Carlo approach. The sensitivity and uncertainty analyses revealed that temperature, atmospheric pressure, protein and lipid contents are parameters sensitive to emission rates, and meteorological parameters have significant impacts on the uncertainty. The established ANN models for the prediction of emission rates can provide scientific evidence and an approach to assess and control the odor and health risk in waste sectors.