Study On Real-time Optimization Methods For Production Processes In An Uncertainty Environment

Today,a plant decision hierarchy is widely-used in industry processes.Each layer in the hierarchy may take different concerns and impact different departments in the plant.The layers work together to optimize the production.This work focuses on real-time optimization(RTO)and model predictive control(MPC)in the hierarchy as well as the integration of the two layers.MPC is for dynamic control including setpoints and reference tracking.RTO will receive variations of the system caused by the uncertainty production environment and then give a response for the variation based on its analysis.The result of the RTO is to update system models and recalculate the economic optimal setpoints,which will be sent to MPC for dynamic control.The thesis opens with modeling for esterification reaction of producing ethyl acetate aided with simulations on Aspen Plus.The model is well suited for the purpose of studying control theories and related technologies.MPC algorithms are implemented on the model.An economic criterion is proposed for optimization as well.Time-varing parameters of plants lead to the model-plant mismatch.Therefore,the parameters of the plant have to be well-estimated.Otherwise,the economic performance will inevitably degrade.When the valid value of the parameter is countable,the online parameter estimation can be viewed as a classification problem and a support vector machine is used to solve it.When the value of the parameter is in a range,a Gaussian regression model is employed to estimate the mean value of the parameter.In addition,an algorithm for detecting the steady state is proposed.For the system with a time-varying economic cost function,a framework embedded with a lookup-table in the RTO layer is developed.The lookup-table which is calculated offline includes candidate steady-state points.When the economic criteria is varied,we search a sub-optimal point in the lookup-table as a temporary setpoint for MPC.The framework is able to avoid the economic losses when the MPC is waiting for the solutions of RTO.Moreover,an algorithm is proposed to recalculate the optimal point which is better to balance the feasibility of MPC and economic-optimality of RTO.When the systems have a relatively long transient process,an approach is developed to calculate a dynamic reference trajectory to improve the economic performance from initial states to the optimal points.It can be viewed as the shortest path problem in the graph theory.The required graph can be obtained by mesh discretization of the state and input space.Thus,the continuous dynamics of the system is described by a finite directed graph represented by a one-step adjacency matrix and relevant weighting matrix.The shortest path problem is solved by Dijkstra's algorithm.Moreover,a sub-graph is extracted at each sampling time to accelerate the computing procedure for online application.Real-time optimization plays an important role in the plant decision hierarchy.It is able to get rid of the uncertainty information since the production environment is varied.RTO techniques interpret them by updating the model and then calculate a new reasonable setpoint for MPC.The integration of RTO and MPC is also a critical problem to guarantee the economic performance.