Application of Real-time Optimization with Persistent Parameter Adaptation (ROPA) to processes using online parameter estimation.

• Ano de defesa: 2020
• Autor(a) principal: Carneiro, Andreia Abadia Borges
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Dados transientes; Dynamic plant; Estimativa em linha; Extended Kalman Filter (EKF); Filtros de Kalman; Online estimation; Otimização em Tempo Real (RTO); Planta dinâmica; Real-time Optimization (RTO); Transient data
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/3/3137/tde-21012021-105517/

Recently, there has been an increasing growth in the optimization processes in the industrial area. The reduction of costs, improvement in the quality of final products and minimization of the environmental risks are important issues that companies must take into consideration. Thus, the development of optimization tools to efficiently identify problems has become suitable. In this context, real-time optimization (RTO) methodology is widely used in industrial area to optimize a plant economically. This is a well-established approach to create a link between a regulatory control and the economical optimization of a process under control. There are several RTO methods which can be used in the optimization cycle. The standard RTO method, called Model Parameter Adaptation (MPA), is one of the most applied in industry. Albeit a good method, there are some problems related to the MPA as well as other RTO methods, such as the use of steady-state (SS) data to update SS models to a dynamic plant, the delay in the detection of the SS condition in the system to start the optimization cycle, and the difficulty to model a complete unit since those methods require it. Real-time Optimization with Persistent Adaptation (ROPA) is a new methodology which tackles those issues. ROPA uses transient data to update the model aiming to optimize the plant. Thus, there is no need to wait for the SS condition because the dynamic plant is not updated with stationary information. Aiming to verify the advantages of this new method, this work presents the results of the ROPA application to two chemical processes. All simulations are performed using MATLAB, the dynamic model and the sensitivity equations are solved by SundialsTB. For the first case study, the Williams-Otto reactor, random and deterministic disturbances are considered in the system in order to simulate a real plant. In addition, the Extended Kalman filter (EKF) is used as the online estimator to obtain the estimated parameters and states in the current time. Regarding the Williams-Otto reactor study, the state estimate results show that the filter works consistently, and the state covariance matrix is satisfactorily tuned. Additionally, the parameter estimation shows that ROPA is able to respond to the disturbances occurrence reproducing the actual plant parameter profile. ROPA runs the economic optimization continuously independently of the plant condition. A Monte Carlo analysis of benefits in applying ROPA method in the RTO cycle shows that the method is suitable to track the plant optimum. Regarding the second case study, the Propylene Chlorination process simulated in a commercial dynamic simulator is optimized by an external ROPA implemented in MATLAB. In this case, ROPA can also reach the stationary optimum, and the filter works properly. However, the MPA and ROPA results are similar because the process is in a gas-phase with fast dynamics. Even in this situation, it can be seen that MPA still has the steady-state delay issue. .