Economic Performance Assessment And Optimization For Multi-Layer MPC

As the most typical Advanced Process Control (APC) technology, Model Predictive Control (MPC) can enhance enterprises economic benefits effectively and has found wide applications in process industry, including petroleum industry, chemical industry, papermaking, pharmaceutical industry, etc. Since the implementation of MPC control system needs additional investment which generally involves a large cost, the benefit and cost ratio is one of the most important factors to be considerated by enterprises. Generally, most MPC controllers operate reasonably well at the early stage. However, their performance may deteriorate after a certain period due to the variation of operating conditions and producion environments. In the worst case, the process might even become uncontrollable and the MPC controllers have to be switched off. resulting in a large waste of investment and loss of economic benefits. Therefore, the economic performance assessment of MPC is important and has attract a lot of interest by academia and industry. Currently, the most common assessment method is the Minimum Variance Control (MVC) benchmark. By comparing the actual variance of the process with the MVC benchmark, the performance improvement potential is provided. However, MVC only deals with process output variance without considering the actuator constraints, so this benchmark is usually deemed as an ideal performance assessing guideline but seldom-implemented in industrial applications. Instead, to implement the economic performance assessment and optimization for multi-layer MPC structure, the LQG benchmark which considers both the manipulated variables variance and the controlled variables variance with a quadratic dynamic index is adopted in this thesis. The research topics include:1.The economic performance assessment is considered for multi-variable controllers in petrochemical processes, an integrated two-layer structure of LQG economic performance steady state optimization and MPC dynamic optimization is given, the relationship between the controlled variables variance and the manipulated variables variance is calculated and accordingly its optimal Pareto surface function is obtained by regression method, and a pragmatic economic performance assessment and optimization strategy for multi-variable process is introduced. Hence the optimal setpoints for MPC and optimial back-off values can be determined.2. An equigrid LQG benchmark is introduced to improve the traditional LQG benchmark which was determined from an asymmetrically distributed discrete points set with unnecessary computation and unsatisfied regression effect. Three approaches consisting of numerical equigrid algorithm, recursive equigrid algorithm and analytical equigrid algorithm for calculating the equigrid LQG benchmark are introduced based on the ARMAX model processes, hence the regression effect is improved and the computing cost is reduced. Based on the LQG benchmark, the sensitivity analysis strategy for each pair of manipulated variable and controlled variable is introduced.3. The LQG benchmark for multi-model processes is introduced to deal with the multiple operation situations caused by materials varying, environment fluctuation, product switching, operator change and so on. According to different causations, the multi-model situations include multiple disturbance models and multiple process models. Based on the analysis of MVC benchmark, the LQG performance assessment and optimization for multi-model processes are introduced.4. To avoid the problem that the calculated optimization setpoints from upper layer are not available in lower-layer because of actuator physical limits, which will lead to control performance deterioration, and to ensure the constraints consistency in multi-layer MPC control and optimization structure, the back-calculation from basic control layer to upper layer is introduced, the consistency of constraints in different layers is proved. The constraints consistency of three typical cascade control loop or transparent control loop of LQGO—PFC—PFC, LQGO-PFC-PID, LQGO-QMPC—PID are analyzed. Based on the above consideration, a cascade level control system is designed and the control results show the effectiveness of the proposed approach.5. The integrated structure of LQG based steady state optimization and MPC dynamic control is used for economic performance assessment, to replace the current strategies for which the steady state optimization and dynamic control are separated. The crucial variables are selected as the connections between the two layers, to ensure the calculated optimal setpoints from upper layer are always practical in lower layer. The economic performance assessment and optimization for the outlet Oxygen content control loop and the outlet temperature control loop of delayed coking furnace in a large petrochemical enterprise of China are implemented, hence the profit margin in delayed coking furnace is analyzed and the parameters designing rules of MPC controller are given.