Data-based Optimal Control And Its Application In GGBS Production Process

With the development of science and technology,the requirement for high control performance is needed more urgently.However,high nonlinear and unknown dynamics are always existed in complex systems,such as chemical process,traffic system and industry manufacture process,therefore,it is usually hard or impossible to build the precise mechanism models.For these kinds of systems,traditional adaptive control methods usually cannot achieve satisfied control consequence.With development of sensor and communication technology,massive data containing large amount of system information is obtained and stored in real time.By using process data rather than mechanism model,data-based optimization and control method has been tested an effective way to realize the control of complex nonlinear system.Combing neural network,adaptive critic design,reinforcement learning and traditional dynamic programming,adaptive dynamic programming(ADP)effectively avoids the 'dimension disaster' problem and is recognized as a successful optimal control method for nonlinear system.Driven by process data,further research on adaptive dynamic programming theory and corresponding application to ground-granulated blast-furnace slag(GGBS)production process shows important theoretical and application value.In this dissertation,based on process data and adaptive dynamic programming method,optimal control problems(setting value optimization,overshoot decrease,control constraints and controlled plant with jumping parameters)are investigated,and corresponding theorems are applied to GGBS production process.The main research of this dissertation can be briefly described as follows:(1)For a kind of discrete-time nonlinear system with control constraints,by introducing a non-quadratic index function and transforming the tracking problem into a regulation problem,the trajectory can be tracked in the optimal way with corresponding control input guaranteed in given range.Further,above ADP tracking method is improved to realize optimal tracking control with control constraints when system model is partly known.Based on the idea of multiple model adaptive control,a new multiple set-points tracking control strategy is given to decrease overshoot controller.Given trajectory is divided into multiple setpoints,and ADP controller is designed to track the trajectory step by step.With guaranteed stability,dynamic response and control performance are obviously improved.(2)For nonlinear system with jumping parameters,an ADP controller on the basis of multiple model adaptive control scheme is proposed.Multiple submodels are built to cover system uncertainty,at the same time,multiple subcontrollers are designed based on proposed ADP method.Switching mechanism is introduced to decide the closest model to current system and the nearest initial parameters,so that the initial admissible condition is always satisfied for the whole time.Optimal tracking control is realized for system with jumping parameters and good control performance is guaranteed.(3)In the case that safe operation is ensured,and production quality and yield reach the requirement,a plant-wide optimization strategy considering both total production effectiveness and economic benefit for GGBS production process is proposed.Based on massive process data,a LS-SVM algorithm based on PSO optimization is proposed to establish the models of production yields,quality and grinding pressure difference in GGBS production process.The multiple objective problems of GGBS production process is formulated and solved by evolutionary algorithms,and corresponding optimal solution is given as optimal set value.(4)After obtaining dynamic model built by recurrent neural network utilizing process data,proposed ADP method with control constrains is successfully applied to GGBS production process.By analyzing massive process data,typical working conditions are extracted.Further,a multiple model ADP controller is designed to realize the optimal control of GGBS production in multiple conditions.