Modelagem do tratamento de efluentes contendo ciprofloxacina via ozonização com aplicação da técnica de redes neurais artificiais
Data
2022-10-26
Autores
Parise, Barbara de França 
Orientadores
Moraes, José Ermírio Ferreira de
Tipo
Dissertação de mestrado
Título da Revista
ISSN da Revista
Título de Volume
Resumo
Devido à persistência dos fármacos no ambiente aquático e a sua resistência aos tratamentos
convencionais aplicados aos efluentes industriais e domésticos, tornase necessário a busca por
novos processos capazes de degradar tais poluentes. Nesse contexto, surgiram os processos
oxidativos avançados (POA) que utilizam o radical hidroxila para degradar os poluentes
orgânicos, podendo mineralizálos à água e gás carbônico. Os POA podem ser classificados
como químicos ou fotoquímicos e como homogêneos ou heterogêneos, podendo também
envolver o uso de outras tecnologias associadas como o uso de microondas (processos
sonoquímicos) e de energia elétrica (processos eletroquímicos), por exemplo. Os POA, de
forma geral, resultam em um complexo mecanismo de reações químicas, envolvendo a
participação de radicais livres. Além disso, em diversos casos, ocorre a aplicação da radiação
ultravioleta (UV) e visível (Vis) nos processos. Dessa forma, a modelagem fenomenológica dos
POA tornase bastante complexa. Neste contexto, a aplicação da técnica das redes neurais
artificiais (RNA), que são formas de modelagens empíricas, baseadas no mecanismo de
aprendizagem do cérebro humano, podem ser uma importante alternativa para modelar estes
processos. O presente trabalho teve como objetivo modelar, usando a técnica das RNA, o
processo de degradação do antibiótico ciprofloxacina (CIP) por ozonização. O modelo
desenvolvido apresentou as seguintes variáveis de entrada: (i) tempo, (ii) pH, (iii) concentração
de ozônio na corrente de entrada, (iv) concentração inicial de ciprofloxacina. Como variáveis
dependentes (de saída), foram adotadas o teor de carbono orgânico total (COT) e a concentração
de ciprofloxacina ([CIP]). Foram desenvolvidos 2 modelos distintos, sempre apresentando uma
camada de entrada com 4 variáveis, uma camada oculta e uma camada de saída com uma
variável, sendo o número de neurônios da camada oculta devidamente otimizado. Para o modelo
1, definiuse o COT como a saída da rede e o [CIP] como saída para o modelo 2. A melhor
configuração de rede obtida para o modelo 1 apresentou uma camada oculta com 7 neurônios,
um coeficiente de determinação (R2
) igual a 0,9939 e função objetivo igual a 0,0035 para o
treinamento. Para o modelo 2, a melhor configuração da rede obtida foi com o uso de 9
neurônios na camada oculta, que apresentou melhores valores de coeficiente de determinação
e função objetivo para o conjunto treinamento (R² = 0,9966 e função objetivo = 0,0016),
resultando em um bom ajuste entre os dados preditos e experimentais. Além disso, as duas redes
neurais obtidas foram também comparadas com modelos empíricos polinomiais. Para tanto,
foram realizadas simulações da RNA1 e da RNA2, para a previsão do COT em 30 min e da
[CIP] em 2 min de reação, considerando todas as condições experimentais adotadas no presente
trabalho. Os resultados dessas simulações foram então confrontados com os dados
experimentais, calculandose os valores de R
2
. O modelo RNA1 apresentou um R
2 de 0,9931,
enquanto que o modelo polinomial apresentou um R
2 de 0,9384. No caso do modelo RNA2, o
mesmo apresentou um R
2
igual a 0,9977 e o modelo polinomial um R
2 de 0,9871. Dessa forma,
verificouse que os dois modelos, obtidos via redes neurais, apresentaram uma melhor
capacidade de predição do que os modelos polinomiais. De forma geral, foi possível concluir
que processo de degradação de ciprofloxacina, via ozonização, foi modelada com sucesso pela
técnica das redes neurais artificiais, sendo capaz de predizer, de forma satisfatória a remoção
de COT e a degradação da CIP, considerando o domínio experimental adotado.
Due to the persistence of drugs in the aquatic environment and their resistance to conventional treatments applied to industrial and domestic effluents, it becomes necessary to search for new processes capable of degrading such pollutants. In this context, the advanced oxidative processes (AOP) that use the hydroxyl radical to degrade organic pollutants have emerged and can mineralize them to water and carbon dioxide. AOPs can be classified as chemical or photochemical and as homogeneous or heterogeneous and may also involve the use of other associated technologies such as the use of microwaves (sonochemical processes) and electrical energy (electrochemical processes), for example. AOPs, in general, result in a complex mechanism of chemical reactions, involving the participation of free radicals. In addition, in several cases, ultraviolet (UV) and visible (Vis) radiation processes occur. Thus, the phenomenological modeling of POA becomes quite complex. In this context, the application of the technique of artificial neural networks (ANN), which are forms of empirical modeling, based on the learning mechanism of the human brain, can be an important alternative to model these processes. The present work aimed to model, using the ANN technique, the degradation process of the antibiotic ciprofloxacin (CIP) by ozonation. The model developed presented the following input variables: (i) time, (ii) pH, (iii) ozone concentration in the input stream, (iv) initial ciprofloxacin concentration. As dependent (output) variables, total organic carbon (TOC) content and ciprofloxacin concentration ([CIP]) were adopted. Two distinct models were developed, always presenting an input layer with 4 variables, a hidden layer and an output layer with one variable, with the number of neurons in the hidden layer being properly optimized. For model 1, TOC was defined as the network output and [CIP] as the output for model 2. The best net configuration obtained for model 1 had a hidden layer with 7 neurons, a coefficient of determination (R2 ) equal to 0.9939, and an objective function equal to 0.0035 for training. For model 2, the best net configuration obtained was with the use of 9 neurons in the hidden layer, which presented better values for the coefficient of determination and objective function for the training set (R² = 0.9966 and objective function = 0.0016), resulting in a good fit between the predicted and experimental data. In addition, the two neural networks obtained were also compared with empirical polynomial models. To this end, ANN1 and ANN2 simulations were performed to predict TOC in 30 min and [CIP] in 2 min of reaction, considering all the experimental conditions adopted in the present work. The results of these simulations were then confronted with the experimental data, calculating the R2 values. The RNA1 model presented a R2 of 0.9931, while the polynomial model presented a R2 of 0.9384. In the case of the RNA2 model, it presented a R2 equal to 0.9977 and the polynomial model a R2 of 0.9871. Thus, it was verified that the two models, obtained via neural networks, presented a better prediction capacity than the polynomial models. In general, it was possible to conclude that the degradation process of ciprofloxacin, via ozonation, was successfully modeled by the artificial neural networks’ technique, being able to predict, in a satisfactory manner, the removal of TOC and the degradation of CIP, considering the experimental domain adopted.
Due to the persistence of drugs in the aquatic environment and their resistance to conventional treatments applied to industrial and domestic effluents, it becomes necessary to search for new processes capable of degrading such pollutants. In this context, the advanced oxidative processes (AOP) that use the hydroxyl radical to degrade organic pollutants have emerged and can mineralize them to water and carbon dioxide. AOPs can be classified as chemical or photochemical and as homogeneous or heterogeneous and may also involve the use of other associated technologies such as the use of microwaves (sonochemical processes) and electrical energy (electrochemical processes), for example. AOPs, in general, result in a complex mechanism of chemical reactions, involving the participation of free radicals. In addition, in several cases, ultraviolet (UV) and visible (Vis) radiation processes occur. Thus, the phenomenological modeling of POA becomes quite complex. In this context, the application of the technique of artificial neural networks (ANN), which are forms of empirical modeling, based on the learning mechanism of the human brain, can be an important alternative to model these processes. The present work aimed to model, using the ANN technique, the degradation process of the antibiotic ciprofloxacin (CIP) by ozonation. The model developed presented the following input variables: (i) time, (ii) pH, (iii) ozone concentration in the input stream, (iv) initial ciprofloxacin concentration. As dependent (output) variables, total organic carbon (TOC) content and ciprofloxacin concentration ([CIP]) were adopted. Two distinct models were developed, always presenting an input layer with 4 variables, a hidden layer and an output layer with one variable, with the number of neurons in the hidden layer being properly optimized. For model 1, TOC was defined as the network output and [CIP] as the output for model 2. The best net configuration obtained for model 1 had a hidden layer with 7 neurons, a coefficient of determination (R2 ) equal to 0.9939, and an objective function equal to 0.0035 for training. For model 2, the best net configuration obtained was with the use of 9 neurons in the hidden layer, which presented better values for the coefficient of determination and objective function for the training set (R² = 0.9966 and objective function = 0.0016), resulting in a good fit between the predicted and experimental data. In addition, the two neural networks obtained were also compared with empirical polynomial models. To this end, ANN1 and ANN2 simulations were performed to predict TOC in 30 min and [CIP] in 2 min of reaction, considering all the experimental conditions adopted in the present work. The results of these simulations were then confronted with the experimental data, calculating the R2 values. The RNA1 model presented a R2 of 0.9931, while the polynomial model presented a R2 of 0.9384. In the case of the RNA2 model, it presented a R2 equal to 0.9977 and the polynomial model a R2 of 0.9871. Thus, it was verified that the two models, obtained via neural networks, presented a better prediction capacity than the polynomial models. In general, it was possible to conclude that the degradation process of ciprofloxacin, via ozonation, was successfully modeled by the artificial neural networks’ technique, being able to predict, in a satisfactory manner, the removal of TOC and the degradation of CIP, considering the experimental domain adopted.