Research On Modeling And Advanced Control For Multivariable Thermal Systems Based On Process Data

Most thermal systems are complex multivariable systems with nonlinearity,coupling and constraints on variables.Modelling and control for thermal systems are always hot spots of research in recent decades.Research on identification using operation data and advanced control for thermal power plants are of significant value both in theory and practice.This dissertation mainly focus on the study of identification using operation data and advanced control for thermal systems.Identification approaches for input-output models and state space models are studied and applied into the modelling of coordinated control systems(CCSs)in coal-fired power plants and of organic Rankine cycle(ORC)based waste heat energy conversion systems(WHECSs).Different model predictive control strategies are designed based on input-output data and identified models for a CCS of a coal-fired power plant and an ORC based WHECS.The main work of this dissertation are summarized as follows:1.The essence of system identification is an optimization procedure using input-output data.An improved estimation of distribution algorithm(EDA)with extreme elitism selection and quasi-opposition based learning is proposed as the system identification method.Since the feasible region's convexity and the objective function's continuity is unrestricted,this method is of wider applicability than traditional optimization methods.Improvements are about selection mechanism and search mechanism.The introduction of extreme elitism selection and quasi-opposition based learning,increases the converge speed while taking the population diversity into account.Simulation test results on benchmark functions show the excellent converge speed and global search ability of the improved method compared with other intelligent optimization algorithms.2.For the identification criterion,a hybrid approach combining minimum error entropy(MEE)criterion and the improved EDA method is developed for system identification.The MEE criterion is introduced into system identification to expend its scope to nonlinear or non-Gaussian systems.The improved EDA with extreme elitism selection and quasi-opposition based learning has an excellent converge speed and global search ability for the optimal parameters.Simulation test results on benchmark systems demonstrate the efficacy under nonlinear conditions.Then,the approach is applied into identifications of a CCS in a coal-fired power plant using operation data and of an ORC based WHECS using step disturbance experiment data.Simulation results show the effectiveness of the identification method for industrial thermal systems.3.An improved closed loop subspace identification method using innovation estimation is proposed,since many closed loop subspace identification methods are too complicated for large-scale industrial processes.State space matrices are acquired by simple matrix operations such as element extraction from a matrix.Procedures like stochastic realization and auxiliary variables'acquisition are omitted.The proposed method is applied into identification of a CCS in a coal-fired power plant using operation data.Simulation results illustrate it outperforms open loop subspace identification under closed-loop conditions.4.Taking model mismatch into account,a data-driven closed-loop subspace predictive control strategy is developed.It is a combination of closed loop subspace identification and model predictive control method.It leaves out solve procedure of state space models,and predicts system outputs using input-output data.The method is applied into a CCS in a thermal power plant.Simulation results demonstrate its effectiveness and superiority on set-point tracking performance and system robustness.5.A multiple model predictive control strategy using identified models is applied into an ORC based WHECS.The rotating speed of the pump and the shaft torque of the expander are used to represent the operating conditions.Taking into account constraints on manipulated variables,a multiple model predictive controller is designed.Simulation results show that the controlled system can operate well under different operating conditions,and confirm the efficacy of the proposed control scheme.