Análise da dispersão de poluentes em rios via aproximação Bayesiana

• Ano de defesa: 2019
• Autor(a) principal: Faria, Ruan de Rezende
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Química; UFES; Programa de Pós-Graduação em Engenharia Química
• Palavras-chave em Português: Modelagem ambiental; Filtro de partículas; Problemas Inversos; Suavizador de Kernel; Métodos heurísticos; Environmental modeling; Particles filter; Inverse problems; Kernel smoothing; Problemas inversos (Equações diferenciais); Modelagem; Engenharia Química; 66.0
• Link de acesso: http://repositorio.ufes.br/handle/10/11015

Study of the dispersion of pollutants in rivers is related to the analysis of the load of conservative pollutants released instantaneously (diffuse sources of pollution) or of continuous way (point sources) in a given river. These problems were addressed in this master's dissertation by means of experimental data with conservative tracer injected in pulse and continuously. The objective was to identify tracer characteristics (i.e., magnitude, spatial distribution and duration of activity). The approach to the dynamics of these problems was carried out by means of approximation of inverse problems, being computed by the Simulated Annealing (SA) method and the assimilation of data based on the Bayesian methods of Particle Filtering (PF). The SA method allowed obtaining satisfactory results in the representation of the dynamics of the system off-line. These results were statistically evaluated by using 120 algorithm repetitions and different levels of uncertainty. However, one important limitation was the computational time to approximate on-line. For this reason, the PF (particle filtering) has been used to quantify the uncertainties in different conditions of priori distribution and number of particles, observing sequentially the performance of the SIR and SIR particle filter with Kernel smoothing in relation to the Root Mean Square (RMS), coefficient of determination (R2 ), autocorrelation coefficient of the residue (SC) and minimum resampling coefficient (REAmin). Better results were obtained through the Kernel smoothing technique, responsible for reducing sample depletion of the SIR particle filter. Therefore, the use of the SIR filter with the Kernel smoothing has been confirmed as an alternative for possible applications of river pollution control, virtual inference and alarm.