Study On Multi-model Control And Its Applications For Nonlinear Multivariable Thermal Processes

With the fast development of modern society and economy, on the one hand technological processes of industrial production become more and more sophisticated, and processes with properties of strong nonlinearity, parameter time-varying, and high dimensions are constantly emerging. On the other hand, in order to meet the demand of energy conservation and improving production quality, the requirement for control performance is also getting higher and higher. Therefore, due to decreasing control effects for nonlinear and strong interaction multivariable processes, conventional single loop control structures and algorithms designed based on linearization model of steady points become unable to adapt these challenges. However, because of difficulties in modeling, control structure and controller tuning not easy to deal with global linearity and interactions, nonlinear multivariable control system design is a sustainable problem not only in theory but also in engineering practice.In thermal industries, many typical processes, e.g., ball mill coal-pulverizing system (BMCPS), fossil fuel power unit (FFPU), etc., are belongs to the nonlinear multivariable system with strong interactions. Moreover, these processes usually run in a very wide-range work conditions which make their control more complex. Despite of many advanced process control (APC) algorithms in recent years, APC has very limited engineering applications on account of lack of software and hardware support, effective relationship of controller parameter and plant physical significance. In direct control level of practice process control, the PID algorithm is also keep dominant in loop control. Consequently, in order to meet the demands of high and reliable economic operations, study on available and practical control strategy based on PID algorithm for MIMO thermal systems has a enormously significance.On the background of the typical thermal processes, using multi-model method to deal with nonlinearity and operation point varying, and using PID control structure and appropriate parameter tuning to tackle loop interactions, a hierarchical structure with a direct control layer and a supervisory-optimization layer is constructed, and a systematical design method for nonlinear multivariable process is investigated. The overall strategy is suitable for industrial process control in a wide-range operation. The main research contents and innovational work of this thesis are concluded as follows.(1)As for modeling and identification of nonlinear MIMO processes, a control oriented modeling method is provided, and a frequency-domain based model parameter identification approach is discussed. A control oriented BMCPS model is structed by mechenism analysis, parameter identification, and heuristic simplization, which provides a simulation platform for the BMCPS control system design. To get the local models for multi-model control, a step sequence test method is employeed, then closed loop system is divided into several SISO opened-loop systems, and a min-max searching algorithm in frequency domain is used for computing the transfer functions for all individule loops. The control oriented modeling method provide a effective and simple strategy for control system design, and the frequency-domain-based MIMO closed-loop identification gives a approach to high accurate local models for nonlinear MIMO processes.(2)As for local control structure of the multi-model framework, a simple and available algorithm for control configuration selection is proposed. Firstly, five basic PID control structures are summarized up: decentralized, centralized, sparse, full decoupling, and partial decoupling control, which have different sturcture characteristics, respectively. Then through interaction analysis for MIMO processes base on relative normalized gain array (RNGA) and interation index array (IIA), a selection critiren and an algorithm for control structure are proposed. This method can be applied to MIMO systems with integrating properties and non-square systems. Interaction index can be directly and simply calculated through the process transfer function matrix, and it can be easily used for industrial applications. The control structure selection provides the foundament for good design of PID controllers. Simulation results illustrate the effectiveness of the proposed algorithm of control configuration selection.(3)As for local controller design, the method of controller design and parameter tuning for the basic five PID control structures are given respectively. Firstly, based on the conception of effective transfer function (ETF), a systematical method for designing decentralized, centralized, and spase controller is given, which make the controller design just as designing individual SISO control loops, and not regarding interaction among process chanels. Then for full decoupling and sparse decoupling (partial decoupling), based on an aproxmated inverse matrix of process transfer funtion using power spectrum and decoupling index, an analytic design method of a decoupling compensator matrix is proposed. The compensator matrix gives consideration to desired closed-loop performance, and easy to be computed and to be implemented in industrial applications. In addition, a factorization method is porposed to deal with controller design for interating MIMO processes. Particularly, a novel DMC-based PID control strategy is proposed for long time-delay process to deal with slow output responses and improve control performance, which combines both merits of the PID and MPC algorithm. Many simulation results shows that the proposed systematic controller design method is very effective and simply.(4)As for the design of supervisory and optimization layer in the multi-model control framework, an integrated solution is given through using two typical thermal processes as examples. Considering the nonlinear properties of BMCPS and fossil fuel power unit (FFPU), a multi-model division strategy is proposed in order to deal with the problem for local model divisiong and switching. A fuzzy swithing approach is applied to determine weighted controller output values of every local model, and realize a stable bumpless switching control. Considering the properties of the varying work conditions, performance assessments should be employeed so as to add extra local models and controllers if necessary. As for setpoint optimization in the supervisory control layer, a method based on steady-state gain to determine initial optimization value is designed, which can be easily extended to be a self-optimization way as changing processes characteristics. In the proposed multi-model hierarchical structure, the direct control layer is to deal with loop control for setpoint tracking and disturbance rejection, and the supervisory-optimal layer is to schedule controllers and optimize loop setpoints. The two layers cordinate each other to achieve control aims. Simulations for BMCPS and FFPU illustrate the effectiveness of the proposed control architarctue.(5)The proposed multi-model control framework is successfully applied to a BMCPS control engineering project. Base on simulation and improvement using the constructed BMCPS model, and considering the BMCPS chracteristics and control requirements, hardware configuration, control strategy design, software realization, and engineering implement are carried out in sequence. The software packages for setpoint optimization, multi-model switching, process supervisory, and PLC process control are all developed. Application in industry field shows that the proposed BMCPS cotrol system can not only make dynamic responses better, improve the adaptation for varying work condidtion, and get better control performance, but also decrease unit electrcity comsumption for coal pulverizing, and energy conservation can be achieved.Finally, the work conclusions and major contributions of this thesis are summarized, and some possible future research directions are provided.