A Study On Multi-Model Based Advanced Control Approaches For Thermal Processes

As the thermal power units grow in size and participate in grid load regulation more frequently, the thermal processes reveal much more complex dynamics, such as nonlinearity, coupling among the multitude of variables, large time-delay, tight operating constraints, and unknown disturbances. Consequently, the conventional linear control approaches are no longer sufficient in meeting performance specifications. Given these reasons, this thesis proposes to use multi-model strategy to develop suited control oriented models for the nonlinear power plants. Then the advanced predictive control and sliding mode control approaches based on multi-model are designed for the complex thermal processes to improve the control performance of the power plants.The main contributions of this thesis are as follows:(1) By taking local model network as the expression form of nonlinear system, a modified multi-model modeling method on the basis of satisfactory fuzzy clustering technique is proposed. Unlike the conventional clustering method, whose performance is greatly relying on the pre-specified clustering centers, we propose to initialize the clustering centers according to the singular value decomposition of covariance matrix of sample, and determine the clustering number through the combination of accuracy requirement and cluster validity index. The structure of the local model network (LMN) can then be determined directly from the results of the clustering, and weighted performance function based identification algorithm is finally used to estimate the local model parameters. Simulation results of Bell-Astrom boiler turbine coordinated system show that the proposed modeling method can achieve high identification accuracy with less number of local models and lower computational consumption.(2) Two multi-model predictive controllers based on the local model network (LMN) are proposed, i.e. the controller-weighted multi-model predictive controller (CWMMPC) and the model-weighted multi-model predictive controller (MWMMPC). For the CWMMPC, the value of local model’s validity function is directly utilized as the weight of corresponding local controller, and the constraints of feasible region of each local model are taken into account in the design of the local predictive controllers, thus performance degradation caused by the invalidation of local model in global region can be alleviated. For the MWMMPC, the LMN is utilized to develop a time-varying global predictor and the improved immune genetic algorithm (IGA) is devised to calculate the optimal control law at each time instant. By using the local stabilizing controller and terminal cost function, the proposed MWMMPC can guarantee the stability of the closed-loop system while effectively shorten the prediction horizon thus reducing the computational complexity. The advantages and effectiveness of these two controllers are validated through some simulations on Bell-Astrom boiler-turbine system and a 500MW power plant unit.(3) To deal with unknown disturbances in thermal processes, an improved disturbance observer is designed and then is used to form a so-called modified disturbance observer based multi-model predictive controller (MDOB-MMPC). In MDOB-MMPC, the design of local controller is based on compound control technique. More precisely, the estimation of the unmeasured disturbances is used as a feedforward compensation signal to remove the effect of disturbances quickly and is also considered in the MPC framework to adjust the input constraints, while the MPC control law in the feedback channel is designed based on the nominal plant. Simulation results on the superheated steam temperature system show that the proposed control method can adapt to the load changes within a wide operating range rapidly and has a satisfactory unknown disturbance rejection performance.(4) To overcome the nonlinearity of the power plant over wide operating range, and the uncertainty caused by coal variation, environment change and model simplification, a multi-model based predictive sliding mode controller (MMPSMC) is proposed for the thermal processes with large inertia behavior or time-delay. Each local controller in MMPSMC is a predictive sliding mode controller based on second-order plus dead-time (SOPDT) model, which is easily implemental. A predictive model without the delayed output is firstly developed from an identified nominal SOPDT model to estimate the future output, and then reasonable sliding function and sliding mode control law are designed to asymptotically stabilize the closed nonlinear system in the case of uncertainties and disturbances. Simulations results on bed temperature control system of circulating fluidized bed (CFB) boiler show that a desired control performance can be attained within a wide operation range, even in the presence of significant unknown disturbances.