Desenvolvimento de modelos de previsão de desempenho de estações convencionais de tratamento de água

Abstract

This work consisted in the development of two models using the Fuzzy Logic in order to predict the performance of conventional water treatment plants, based on finished and settled turbidity. The Model for Performance Prediction 1 (MPD1 ) was developed based on indices as tools used in making decisions about the process improvements. Due to the inherent limitations of these indices it was also developed the MPD2, taking into account the most important variables related to the raw water quality and the treatment plant characteristics. Both models were developed based on database of eight plants which average flow rate range from 190 L.s-1 to 5,400 L.s-1. The conventional process was chosen because it is the most common water treatment process in Brazil and the Fuzzy Logic usually permits the insertion of the experts' knowledge in the development of indices and models. In addition, it works with data that doesn't have a linear pattern that would become very difficult the use of classical mathematical models. The difficulty to develop a generic model for predicting the finished turbidity was confirmed mainly due to: the great number of interactions in the water treatment process, the raw water quality, the frequency of monitoring of raw and finished water, and the lowest variability of the plant's finished turbidity for compliance with the drinking water quality standards. It was evidenced that there is low correlation coefficients between the finished turbidity predicted by the models and the finished turbidity observed in database. The MPD2 was viable to estimate the median finished and settled turbidity in spite of the limitations to deal with extreme values. Finally, it is possible to infer with use of the models that the raw water quality seems to be more relevant than the characteristics of the treatment plant in order to comply with more restrictive drinking water quality standards.