Azadi Sama, Amiri Hamid, Rakhshandehroo G Reza
School of Engineering, Department of Civil & Environmental Engineering, Shiraz University, Shiraz, Iran.
Department of Civil & Environmental Engineering, Tarbiat Modares University, Tehran, Iran.
Waste Manag. 2016 Sep;55:220-30. doi: 10.1016/j.wasman.2016.05.025. Epub 2016 Jun 2.
Waste burial in uncontrolled landfills can cause serious environmental damages and unpleasant consequences. Leachates produced in landfills have the potential to contaminate soil and groundwater resources. Leachate management is one of the major issues with respect to landfills environmental impacts. Improper design of landfills can lead to leachate spread in the environment, and hence, engineered landfills are required to have leachate monitoring programs. The high cost of such programs may be greatly reduced and cost efficiency of the program may be optimized if one can predict leachate contamination level and foresee management and treatment strategies. The aim of this study is to develop two expert systems consisting of Artificial Neural Network (ANN) and Principal Component Analysis-M5P (PCA-M5P) models to predict Chemical Oxygen Demand (COD) load in leachates produced in lab-scale landfills. Measured data from three landfill lysimeters, including rainfall depth, number of days after waste deposition, thickness of top and bottom Compacted Clay Liners (CCLs), and thickness of top cover over the lysimeter, were utilized to develop, train, validate, and test the expert systems and predict the leachate COD load. Statistical analysis of the prediction results showed that both models possess good prediction ability with a slight superiority for ANN over PCA-M5P. Based on test datasets, the mean absolute percentage error for ANN and PCA-M5P models were 4% and 12%, respectively, and the correlation coefficient for both models was greater than 0.98. Developed models may be used as a rough estimate for leachate COD load prediction in primary landfill designs, where the effect of a top and/or bottom liner is disputed.
在未经控制的垃圾填埋场进行垃圾掩埋会造成严重的环境破坏和不良后果。垃圾填埋场产生的渗滤液有可能污染土壤和地下水资源。渗滤液管理是垃圾填埋场环境影响方面的主要问题之一。垃圾填埋场设计不当会导致渗滤液在环境中扩散，因此，工程垃圾填埋场需要有渗滤液监测计划。如果能够预测渗滤液污染水平并预见管理和处理策略，此类计划的高昂成本可能会大幅降低，且计划的成本效益可能会得到优化。本研究的目的是开发两个由人工神经网络（ANN）和主成分分析 - M5P（PCA - M5P）模型组成的专家系统，以预测实验室规模垃圾填埋场产生的渗滤液中的化学需氧量（COD）负荷。利用来自三个垃圾填埋渗滤计的测量数据，包括降雨深度、垃圾填埋后天数、顶部和底部压实粘土层（CCL）的厚度以及渗滤计上方覆盖层的厚度，来开发、训练、验证和测试专家系统，并预测渗滤液COD负荷。对预测结果的统计分析表明，这两个模型都具有良好的预测能力，其中ANN比PCA - M5P略具优势。基于测试数据集，ANN和PCA - M5P模型的平均绝对百分比误差分别为4%和12%，且两个模型的相关系数均大于0.98。所开发的模型可用于初步垃圾填埋场设计中渗滤液COD负荷预测的粗略估计，在这种设计中，顶部和/或底部衬垫的效果存在争议。
