Inteligência artificial para avaliação da qualidade da água

• Autor(a) principal: Silva, Igor Santos
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFS
• Texto Completo: https://ri.ufs.br/jspui/handle/riufs/14265
• topic: Recursos hídricos; Qualidade da água; Clorofila; Reservatórios; Machine learning; Chlorophyll-a; Water quality; ENGENHARIAS::ENGENHARIA SANITARIA

Water quality is fundamental to the preservation and continuity of ecosystems and is also fundamental in the development of human activities, therefore identifying and understanding the phenomena that occurs in it, as the degradation resulting from this last factor, is fundamental in the study of water resources. In order for this study to attend and serve the diverse stakeholders in the use of water, identifying the physical, chemical and biological parameters and their changes caused by anthropic activities is essential. Nevertheless, this study is not so easy due to the lack of adequate monitoring of water bodies and the scarcity of data that is characteristic of monitoring programs. In this sense, in this crucial quest to understand the impacts of anthropogenic activities on water quality, the use of statistics as well as machine learning techniques are essential and also needed to be further disseminated and encouraged as tools that help decision makers. In this paper, the Artificial Neural Networks, with the aid of the time series, and Random Forest were used, which allowed the forecasting and prediction of the chlorophyll-a concentration, which in high concentrations characterizes a water body as eutrophic due to the excess of nutrients present in it. In this manner, it was possible forecasting chlorophyll-a concentration using Companhia Ambiental do Estado de São Paulo (CETESB) and the United State Geological Service (USGS database, Artificial Neural Networks was used, and results were obtained close to the values of test and train sets, according to the metric analysis using Means Square Error (MSE) and the Root Mean Square Error(RMSE).The Random Forest algorithm was used to predict the chlorophyll-a concentration in reservoirs in the state of Sergipe, and even with a smaller database, the model obtained good results that could be improved if more data were available. Therefore, these algorithms presented accuracy in their results that can allow reduce costs of laboratorial and labor analyzes.