Filtro de partículas aceleração-reponderação em um problema referência da engenharia química

Abstract

Due to the constant computational advances, as well as the development of efficient methods for solving nonlinear problems, it has become interesting the use on-line estimation on nonlinear chemical processes. In this context, the search for more faster and robust methods have become a challenge for researchers, for allows the accomplishment of real-time estimates of unmeasured or infrequently-measured variables, states variables and unknown or time-variant model parameters. Therefore, state estimation aims to use the information available through the process model and the measurements to obtain estimates of system states. This information can be used for monitoring, optimization and process control. For these reasons, the present work is aimed the application the Sampling Importance Reasampling (SIR) and MoveReweighting (AR) filters to the state estimation in an oil reservoir problem and the van der Vusse benchmark. For this, the various estimation methods are discussed, focusing on the particle filters. Next, the two cases proposed in this work were simulated for the application of particle filter algorithms. The results for both filters were satisfactory, but the AR filter obtained better performance. This methodology shows the benets when using a small number of particles and the move-reweighting approach can be promising to attenuate the asymmetry in the distribution of the particle weights, thus obtaining better results for online Bayesian inference and avoiding the need to use the resampling, which is a SIR filter problem.