Aplicação de tecnicas estatisticas multivariadas e de redes neurais na modelagem de um sistema de tratamento de efluentes industriais

AUTOR(ES)
DATA DE PUBLICAÇÃO

2003

RESUMO

One of the most difficult problem in the modeling and control of wastewater treatment processes is the construction of reliable process models. For these processes, the development of detailed models based on fundamental principIes and intense kinetic studies is very difficult, expensive and time consuming. The aerated lagoon is a common example of such an environrnental processo Its inflow is variable (both in quality and quantity); the population of microorganisms varies over time, both in quantity and in the number of species; process knowledge is scarce and the few online analyzers available tend to be unreliable. The amount of organic matter present is measured as either the biochemical oxygen demand (BOD) or the chemical oxygen demand (COD). They are key process variables in water quality. It is very desirable to have a reasonably accurate input-output model for BOD prediction because there is a five-day delay in its laboratory analysis, and a significant hydraulic residence time delay in the aerated lagoon. Recent1y, some papers using artificial neural networks (ANNs) in modeling biological wastewater treatment processes have been published. But surprisingly, very little attention has been paid to the dynamic characteristics of these systems. The main objective of this study is to develop an estimation model that provides accurate predictions ofthe biochemical oxygen demand (BOD) ofthe input and output streams of an aerated lagoon at a pulp and paper mill operated by International Paper of Brazil Steady-state and dynamic predictive models are presented based on both ANNs and linear multivariate regression techniques. Water quality measurements - BOD, COD, flow rate, pH, conductivity, color, temperature together with milling process information - pulp and paper production -, over a four-year period are used to develop the models. The advantages and drawbacks of both neural networks and multivariate linear regression techniques in their ability to model complex and multivariate processes are verified and described in details. The ANN models were slight1y more accurate but both models provide reasonably

ASSUNTO(S)

modelagem de dados aguas residuais redes neurais (computação) industria de celulose analise de regressão

ACESSO AO ARTIGO

Documentos Relacionados