Applying artificial neural networks (ANNs) to solve solid waste-related issues: A critical review. (1st April 2021)

• Record Type: Journal Article
• Authors: Xu, Ankun; Chang, Huimin; Xu, Yingjie; Li, Rong; Li, Xiang; Zhao, Yan
• Is Part Of: Waste management. Volume 124(2021)
• Page Start: 385
• Page End: 402
• Publication Date: 2021-04-01
• Subjects: artificial neural network (ANN) -- Solid waste -- Prediction -- Feedforward neural network -- Model configuration; AD Anaerobic Digestion -- MLPANN Multilayer Perceptron ANN -- AI Artificial Intelligence -- MLR Multiple Linear Regression -- ANN Artificial Neural Network -- MSE Mean Squared Error -- BOD Biochemical Oxygen Demand -- MSW Municipal Solid Waste -- COD Chemical Oxygen Demand -- NRMSE Normalized Root-Mean-Square Error -- GA Genetic Algorithm -- PCA Principal Component Analysis -- GBRT Gradient Boosting Regression Tree -- PCR Principal Component Regression -- GDP Gross Domestic Product -- PLS Partial Least Squares -- GM Grey Model -- R2 R-squared -- GP Genetic Programming -- RBFANN Radial Basis Function ANN -- GRNN General Regression Neural Network -- RF Random Forest -- kNN k-Nearest Neighbour -- RMSE Root-Mean-Square Error -- L-M Levenberg–Marquardt -- RSM Response Surface Method -- MAE Mean Absolute Error -- SOFM ANN Self-Organising Feature Map ANN -- MAPE Mean Absolute Percentage Error -- SVM Support Vector Machine -- MBE Mean Bias Error -- SVR Support Vector Regression -- MIMO Multi-Input Multi-Output -- TS Total Solid -- MISO Multi-Input Single-Output -- VS Volatile Solid; Hazardous wastes -- Periodicals; Refuse and refuse disposal -- Periodicals; 363.728
• DOI: 10.1016/j.wasman.2021.02.029
• Languages: English
• ISSNs: 0956-053X

Abstract

Graphical abstract: Highlight: ANN applications on solving solid waste related issues in last decade are reviewed. Studies are classified into macroscale, mesoscale, meso–microscale and microscale. Various configurations on ANN framework and performance evaluation are compared. Preferable settings and promising application in waste related studies are discussed. Abstract: Artificial neural networks (ANNs) have recently attracted significant attention in environmental areas because of their great self-learning capability and good accuracy in mapping complex nonlinear relationships. These properties of ANNs benefit their application in solving different solid waste-related issues. However, the configurations, including ANN framework, algorithm, data set partition, input parameters, hidden layer, and performance evaluation, vary and have not reached a consensus among relevant studies. To address the current state of the art of ANN application in the solid waste field and identify the commonalities of ANNs, this critical review was conducted by focusing on a modeling perspective and using 177 relevant papers published over the last decade (2010–2020). We classified the reviewed studies into four categories in terms of research scales. ANNs were found to be applied widely in waste generation and technological parameter prediction and proven effective in solving meso–microscale and microscale issues, including waste conversion, emissions, and microbial and dynamic processes. Given the difficulty of data collection in many solid waste-related issues, most studies included a data size of 101–150. For mathematical optimization, dividing the data into training–validation–test sets is preferable, and the training set is supposed to account for ~70%. A single hidden layer is usually sufficient, and the optimal numbers of hidden layer nodes most likely range from 4 to 20. This review is supposed to contribute basic and comprehensive knowledge to the researchers in general waste management and specialized ANN study on solid waste-related issues.