Research On Steady-State Target Calculation And Online Model Identification For Two-Layer Model Predictive Control

"Industry 4.0" and "Made in China 2025" have been put forward one after another,which make the global industry facing increasingly competition.How to improve product quality and reduce production cost under the increasingly severe requirements of environmental protection and production safety is a problem that enterprises must face and solve."Smart Plant" which integrate process control,plant operations and business systems is born in this context.Its primary goal is to optimize the plant assets,and product with near-zero accidents and sustainable environment,health and safety.As an independent level of the industrial process control architecture,two-layer model predictive control can effectively achieve the optimization of safety,economy and control objectives for industrial processes.Therefore,it has been widely used since its inception,and become the most widely used advanced process control method.At present,the model predictive control theory is relatively mature.However,due to the large amount of online calculation,high implementation requirements,long period and difficult maintenance,two-layer model predictive control is seriously restricted in the further application of rapid and complex processes.Research on fast algorithm,operation instruction and other aspects of two-layer model prediction control not only can greatly expand the application field of two-layer model prediction control,meet the demands of the current industrial development for optimization control,but also combine the theory with the practical application more closely,which would play an important role in energy conservation and consumption reduction.The research of this dissertation is application-oriented.Aiming to narrow the gap between the model predictive control theory and its practical application,and to improve the real-time capability,scientificity and usablity of two-layer model predictive control,new ideas and methods for solving relevant bottleneck problems are proposed based on the new development of modern optimization theory and constraint feedback control theory,and the algorithms are verified by typical industrial application cases.The dissertation mainly includes the following aspects:(1).In view of the problem that operators could not accurately determine the cost coefficient of manipulated variables,a steady-state target calculation method based on multiple priorities of manipulated and controlled variables is proposed.This method divides manipulated variables into cost manipulated variables and minimum movement manipulated variables,and gives the concept of cost priority and calculation method of the priority of manipulated variables.When the feasible region exists,the steady-state target is solved step by step based on the priority of manipulated variables.Otherwise,a feasible region is obtained by relaxing the constraint conditions through the multipriorities of the controlled variables,and then the steady-state target is calculated based on the priorities of the manipulated variables.The method is simple in parameter setting and is suitable for practical application of industrial process.Finally,the practicability of the method is verified by simulation.(2).In view of the problem of high computational complexity of two-layer model predictive control,a method of two-layer model predictive control based on off-line table look-up is studied to promote the application of two-layer model predictive control in programmable logic controller or on-chip machine.In steady-state target calculation layer,the steady-state inputs,steady-state outputs and steady-state error corrections at the previous time are taken as independent variables.The steady-state inputs and steady-state outputs at the current time are optimized offline.Then the corresponding look-up table is established,and the online calculation of steady-state target can be obtained by looking up the table.For the case that online table look-up is not feasible,the optimal steady-state targets are calculated by multiparameter linear programming.The effectiveness of the method is verified by complexity,sensitivity,stability,robustness analysis and simulation.(3).In view of the long period problem of model identification and re-identification in the implementation and maintenance of industrial predictive control,a semi-adaptive two-layer predictive control method based on zone control is proposed.Zone model predictive control is introduced to unify the open-loop test and closed-loop control.Uncorrelated APRBS signals are introduced as persistent excitation.The adaptive amplitude and benefit balance coefficient are used to maximize the signal-to-noise ratio,which improves the efficiency of multi-variable model identification/reidentification and ensures production safety,product quality and economic benefits.This method belongs to an online open loop test.It solves the problem of the correlation between the test signal and undetectable noise and the contradiction between sufficient signal excitation and smooth system operation.In addition,in view of model structure design under this framework,a weak-coupling model structure is proposed.The model quality is calibrated by using different confidence interval scales of the model parameters at a determined significance level.By judging the quality of each sub-model,the uncertain or unstable sub-model is removed from the controller model structure or used as feedforward.The model quality is further improved gradually through re-identification.(4).In view of the problem of high energy consumption and high pollution in the process of aluminum smelting,combining the above research contents,an energy efficiency oriented two-layer predictive control method is proposed.The proposed method takes the numerical simulation and optimization control into consideration.In the aspect of numerical simulation,the energy saving space and energy saving mode of the melting furnace are analyzed by offline simulation to reduce the computational complexity.In the aspect of optimization control,the priority strategy is adopted in steady-state target calculation layer to optimize economic benefits according to the different importance of controlled variables.The weak coupling model controller is used in the dynamic optimization layer to improve the system robustness and to simplify the parameters' adjustment.This method was successfully implemented in the F1 aluminum alloy smelting furnace of a company in tianjin.The main variables related to smelting efficiency met the expected target,which denotes that the proposed method effectively improved the combustion efficiency and reduced the production energy consumption.