Aplicação de Redes Bayesianas na Identificação de Perigos em Sistemas de Abastecimento de Água para Consumo Humano: Estudo de caso no Município de Viçosa-Minas Gerais

• Autor(a) principal: Bezerra, Nolan Ribeiro
• Data de Publicação: 2011
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: http://locus.ufv.br/handle/123456789/837
• topic: Planos de segurança da água; Avaliação de risco; Redes Bayesianas; Water safety plans, Risk assessment; Bayesian networks; CNPQ::ENGENHARIAS::ENGENHARIA SANITARIA::SANEAMENTO BASICO::TECNICAS DE ABASTECIMENTO DA AGUA

This paper aims to apply principles of Bayesian networks as a tool to estimate the occurrence probability of hazards in the watershed, the source water uptaking and distribution network, and to apply the Hazard Analysis and Critical Control Points System (HACCPS) to identify hazards at each stage on the water treatment plant, through a case study involving the city of Viçosa (Minas Gerais/Brazil). To assist in the identification of hazards associated with the water supply system within the watershed, caption zone and distribution network was adopted a probabilistic classifier, called Bayesian learning algorithm, which was used in the construction of the Bayesian network. For the Bayesian network modeling, it was used the public domain software UnBBayes, developed by the Artificial Intelligence Group at the University of Brasilia (UnB). In the water treatment plant it was used the Hazard Analysis and Critical Control Points System (HACCP) to identify hazards at each stage of treatment (coagulation / rapid mixing, flocculation, sedimentation, rapid filtration and disinfection). The hazard identification and risk characterization were made from a survey of secondary and primary data in the entire system of water supply. To determine the probability of dangerous events in the watershed, caption zone and distribution network it was necessary to define the probabilities of an a priori distribution for the Bayesian network construction. The determination of the a posteriori probability was obtained from information obtained in the field. The a priori information from the watershed was obtained through an experts consultation following the Delphi method. For the consultation, a questionnaire was designed with the variables classified in hazardous events and sent to working specialists in the fields of hydrology, water resources, water quality, soils, and geoprocessing. The selected variables (land use and soil utilization, water availability, urban and industrial activities, among other soil characteristics as slope, vegetation cover and superficial drainage. The results agreed upon together with experts were used in the probability tables definition for a priori distribution of the Bayesian network construction. The outcomes of the created network application, with the software UnBBayes for determining the probability of dangerous events in the contributing watershed to caption zone, presented the a posteriori probability of 78.53% in occurrence of hazardous events in the watershed to contribute towards change in water quality at the caption zone. This level of danger was classified as high, resulting in a water spring with potentially significant hazardous events that directly interfere in the water quality and therefore the on the water treatment. In the caption zone it was determined the contamination dangers propensity levels associated with the Ribeirão São Bartolomeu basin. The result of the Bayesian classification model application in the caption zone presented a "high" probability (90%) for the microbiological contamination level and a low probability (70% and 99%) for biological and chemical contamination levels, respectively. In the water treatment plant (ETAI/SAAE) the "decision tree" tool application is used to identify the Critical Control Points (CCP) and Points of Cares (PC) for detecting viruses, bacteria and protozoa identified dangers. It was identified the protozoan as PCC danger at all stages of treatment, except for disinfection which was identified as Point of Care (PA). In bacterial and viral dangers it has been identified CCP only for filtration. The result of the decision tree method utilization at each step of the treatment plant water was consistent with the historical result of the treatment plant (ETA-I/SAAE). monitoring. Thus, the HACCP system was presented as a preventive and appropriate tool to identify the hazards involved in water supply. It was able to provide important information, from the standpoint of setting priorities and preventive or corrective action adoption, rather than just referring to the evaluation of final product quality. The Bayesian classification model application results on the distribution network presented a danger propensity to the physical (80%) and hydro (60%) variables, which were considered xix high. The variable about water quality achieved a value considered as low (60%). The Bayesian inference has increasingly found more space for it s application, being en used in many studies on decision making for watershed and water supply system concerns. Finally, even with some study limitations in this paper, it is possible to conclude that the Bayesian inference technique utilization is a tool for hazard identification in water supply systems for human consumption, based on the Water Safety Plans guidelines.