Projeto de sensores virtuais e estudo de algoritmos para estimação online de parâmetros em dados com excitação intermitente

Abstract

This work studies parameter estimators for dynamical data with intermittent excitation. Systems with intermittent excitation occur in contexts in which the estimation is based on historical data or on real-time data collected during normal process operation. In the latter case, the implementation of virtual sensors that are robust to variations inthe system dynamics is a typical application. The main motivation for this work is the oshore oil extraction. First, real-time recursive algorithms with time-varying weighting for the estimation of time-varying parameters are studied. Next, the problem of dual state-and-parameter estimation is addressed and two approaches for the parameter estimation stage areinvestigated. The first approach is recursive and uses least squares algorithms with the time-varying weighting. The second one updates model parameters in batch mod whenever transients are detected in moving data window. To evaluate the algorithms, both simulated and experimental tests are performed. The simulated system is a massspring damper in which the spring constant varies with time. The experimental case study regards the estimation of downhole pressure for oshore oil processes usin historical data. The results suggest that monitoring persistence of excitation during run time to update models make virtual sensors more robust to dynamic variations. In addition, the feasibility study of the virtual sensor design for the estimation of produced oil flow in oil wells was carried out. We used correlation-based analysis tools in the available historical data. However, the data collected for this task does not embodies relevant dynamic information about the system, making the development ofa virtual sensor an infeasible task.