Modelagem neural dinâmica das características físico-químicas do lixiviado de aterro sanitário.

Abstract

Due to the physical, chemical and biological processes that Municipal Solid Waste (MSW) are subjected to during their degradation in landfills, leachate is a byproduct considered extremely complex, due to its toxic composition and variability over the years. However, it is necessary to know and monitor it because all information about the leachate is valuable and contributes to the development of techniques and application of technologies adapted for the treatment of this effluent. Faced with the problem that involves the exposure of those who handle it, the costs associated with monitoring, and aiming at the possibility of reducing the frequency of leachate collections, in addition to helping to speed up responses to tests that can take days, Artificial Neural Networks (ANN ) can be an alternative in obtaining the characteristics of the leachate. Therefore, this work aims to develop neural models that represent the quantitative characteristics of leachate generated in a landfill located in a semi-arid region of Brazil. In which the collection and characterization of MSW with different landfilling ages and monitoring of waste and leachate were carried out. The meteorological conditions near the landfill were also analyzed. Through leachate monitoring data, it was possible to build experimental databases. In search of better performance for ANN, banks with 125 and 1000 data were synthetically generated, showing the same trend as the original data set. To prepare the data for the generation of ANN, each set of output data corresponding to Biochemical Oxygen Demand (BOD), Chemical Oxygen Demand (COD), Total Ammonia Nitrogen was placed on a common scale, through standardization and normalization. (NAT) and throughput. In this research, variations in the number of neurons in the hidden layer (1 to 20), the algorithms used in training the ANN (trainlm, trainbr, trainingd and trainoss), and the hidden layer activation and output functions (purelin, logsig) were tested. , tansig and elliotsig). And as a performance measure, the following were evaluated, following this order of priorities: the coefficient of determination (R²), mean absolute error (MAE), root mean square error (RMSE) and the square root of the normalized mean error (NRMSE). The results of the gravimetric characterization indicated a strong relationship between the fraction of the “other” component, which represented the highest percentage at all ages of landfilling, with the exception of fresh residue, while organic matter was highlighted in the fresh residue, being reduced to over the years of grounding. As for leachate, the Landfill under study presents exclusive conditions and characteristics that vary with the types of waste deposited, and depending on the climatic conditions of the semi-arid region, having an accelerated biodegradation process, low leachate generation rates and a high polluting potential. . By checking the performance metrics, the best ANN were selected for predicting each desired variable. The ANN selected for predicting BOD, COD and NAT had good performance and can be used to analyze the characteristics of the leachate, while for flow prediction, it is recommended to insert more variables that can improve the coefficient of determination (R²) , between the observed and predicted data.