Department of Mechanical Engineering Science, University of Johannesburg, Johannesburg, South Africa.
Department of Mechanical Engineering, Walter Sisulu University, Butterworth, South Africa.
Environ Technol. 2022 Apr;43(11):1634-1647. doi: 10.1080/09593330.2020.1845819. Epub 2020 Nov 27.
The present waste-management system in most developing countries are insufficient to combat the challenge of increasing rate of solid waste generation. Accurate prediction of waste generated through modelling approach will help to overcome the challenge of deficient-planning of sustainable waste-management. In modelling the complexity within a system, a paradigm-shift from classical-model to artificial intelligent model has been necessitated. Previous researches which used Adaptive Neuro-Fuzzy Inference System (ANFIS) for waste generation forecast did not investigate the effect of clustering-techniques and parameters on the performance of the model despite its significance in achieving accurate prediction. This study therefore investigates the impact of the parameters of three clustering-technique namely: Fuzzy c-means (FCM), Grid-Partitioning (GP) and Subtractive-Clustering (SC) on the performance of the ANFIS model in predicting waste generation using South Africa as a case study. Socio-economic and demographic provincial-data for the period 2008-2016 were used as input-variables and provincial waste quantities as output-variable. ANFIS model clustered with GP using triangular input membership-function (tri-MF) and a linear type output membership-function (ANFIS-GP1) is the optimal model with Mean Absolute Percentage Error (MAPE), Mean Absolute Deviation (MAD), Root Mean Square Error (RMSE) and Correlation Co-efficient () values of 12.6727, 0.6940, 1.2372 and 0.9392 respectively. Based on the result in this study, ANFIS-GP with a triangular membership-function is recommended for modelling waste generation. The tool presented in this study can be utilized for the national repository of waste generation data by the South Africa Waste Information Centre (SAWIC) in South Africa and in other developing countries.
目前，大多数发展中国家的废物管理系统都不足以应对固体废物产生率不断增加的挑战。通过建模方法对废物进行准确预测，有助于克服可持续废物管理规划不足的挑战。在对系统内部的复杂性进行建模时，需要从传统模型向人工智能模型进行范式转变。尽管聚类技术及其参数在实现准确预测方面具有重要意义，但之前使用自适应神经模糊推理系统（ANFIS）进行废物产生预测的研究并未调查这些参数对模型性能的影响。因此，本研究调查了三种聚类技术（模糊 C 均值（FCM）、网格分区（GP）和减法聚类（SC））的参数对使用南非作为案例研究的 ANFIS 模型预测废物产生性能的影响。研究使用了 2008-2016 年期间的社会经济和人口省级数据作为输入变量，以及省级废物量作为输出变量。使用 GP 对 ANFIS 模型进行聚类，并采用三角形输入隶属函数（tri-MF）和线性输出隶属函数（ANFIS-GP1），该模型的平均绝对百分比误差（MAPE）、平均绝对偏差（MAD）、均方根误差（RMSE）和相关系数（）值分别为 12.6727、0.6940、1.2372 和 0.9392，是最优模型。根据本研究的结果，推荐使用具有三角形隶属函数的 ANFIS-GP 进行废物产生建模。本研究提出的工具可以由南非废物信息中心（SAWIC）和其他发展中国家的国家废物产生数据存储库使用。
