Research On Dynamic Modeling And Control Of Typical Nonlinear Process In The Supercritical Power Unit

As nerve center of supercritical unit, control system plays an important role in stable, safe and economic operation, while two obvious problems exist in the actual control system. One is that the reheat steam temperature system is difficult to control automatically; the other is slow load tracking performance of coordinated control system. Around two typical nonlinear subsystems related to the problems above, the research about control-oriented modeling and control including modeling methods, improvement of optimization algorithm, design and improvement of control method is carried out. Based on the model built in this paper, the design of predictive control and simulation are researched, in addition, practical engineering application of boiler-turbine coordinated system is carried out. The main research achievements of this thesis are as follows:1. For resolving the problems of the traditional continuous model identification methods based on step response, a continuous model identification method which can identify model order and model parameters is proposed. The initial steady-state values are included in the identification parameter vector together with the model parameters, therefore, the output of the step response can be directly taken as the original identification data which avoids incremental data processing procedure in traditional identification methods. The impact of measurement noise and unsteady initial conditions for incremental data processing procedure is removed. The separation of pure delay in identification model is achieved by the introduction of a first-order integral part. So all the parameters of the continuous model can be obtained directly. The direct differentiation of noisy signal can be avoided by using a filter which can reduce the effect of the measurement noise for model identification. The order of the identified model is determined by combining classic determination factor R2 with AIC criteria. The model of reheat steam temperature system at three load point (high, medium, low) is built based on the identification algorithm.2. For resolving the problems in actual reheat steam temperature system of Supercritical Unit and predictive control algorithm, a multiple model block laguerre function predictive control (MMBLFPC) algorithm is proposed. Model predictive control is used to design the reheat steam temperature control system for solving the high inertia and big delay. A cost function integrated with economic performance is proposed by improving the objective function of predictive control, the quantity of spraying water is reduced and economy of unit is increased. The nonlinear problem is resolved by adopting multi-model control structure. The use of block-structured way provides more time for the calculation of predictive control. The dimensionality of optimization variables is reduced by optimal variable compression with Laguerre network, so that the computing speed of predictive controller is increased. The state space model through the conversion of reduced order data-driven model is taken as internal model for MMBLFPC, therefore the design of state observer and solving of Diophantine equations are avoided. By the simulation of MMBLFPC at three operating points, good control performance and high economy of reheat steam temperature control system based on MMBLFPC can be obtained. This algorithm with fast calculating speed is more suitable for engineering application.3. To address the problems in the traditional optimization algorithm for gray box modeling, a merit-based competition hybrid search optimal algorithm which can deal with various inequality constraints is proposed. By combining constrained teaching-learning-based optimization algorithm, the space region of feasible solution can be quickly determined. The search ability for non-smooth problems in the selected area can be improved by combining complex method. The problems in complex method about heavy computing load and possible failure for optimal objective function with noise can be resolved by adopting generating set algorithm with bigger search step. Compared with other optimization algorithms for gray box model based on test functions with noise and without noise and coordinated system model, the effectiveness and superiority of this optimization algorithm can be proved. The input disturbances tests of coordinated system on an actual unit at three load segments (high, medium, low) are implemented to obtain more large-scale test data. Static and dynamic parameters of the simulation model are obtained by using the optimization algorithm. Finally, a nonlinear simulation model of coordinated system is established which can be used for a wider load range.4. Aimed at the shortcomings of the traditional nonlinear model predictive control based on linearization, an improved Newton-type nonlinear predictive control is put forward. By using state-dependent multi-points linearization model instead of fixed single-point linearization model, the prediction accuracy of the internal model in predictive controller is improved. The better performance than the traditional single-point linearization nonlinear predictive control with long prediction horizon can be obtained through simulation comparison based on the established coordinated boiler-turbine nonlinear model. For the difficult tuning problem of multivariable predictive control, a multi-objective optimization algorithm Goal attainment combined with the grid evaluation performance indexes and traditional performance indexes is adopted in the parameters optimization of the improved Newton-type nonlinear predictive control, as a result, better controller parameters are obtained. To resolve the more computing complexity for the proposed control algorithm, an idea about transforming complex controller calculation into explicit computing by using support vector machine regression is further proposed. Through the simulation studies based on the non-linear model of boiler-turbine coordinated system with transformed explicit SVM operation of improved Newton-type nonlinear predictive control, the effectiveness of proposed algorithm can be proved.5. A simplified multi-model switching predictive control algorithm which can be easily realized by a computer is proposed and used for engineering application in boiler-turbine coordinated system. By selected control horizon 1 and the single predictive coincide point, the calculation of predictive control is simplified. The output prediction computation is decomposed into two parts including free output and force output, which avoids solving the Diophantine equations. A simplified quadratic programming algorithm capable of adjusting computing time is obtained by modifying traditional quadratic programming algorithm. The computing time of predictive control can be reduced by using simplified quadratic programming algorithm. Multi-model switching control strategy is used to resolve nonlinear problem. In order to realize undisturbed switching, the fuzzy membership function is introduced into the performance index so that a new fuzzy soft switching method is proposed with the feature of improved computational efficiency when realizing undisturbed switching. Lots of simulation tests of the proposed algorithm are performed on the simulator of supercritical unit, the effectiveness of this algorithm is proved. At last, this control method is used to optimize coordinated control system in an actual thermal power plant. Better performance is obtained after optimization. The stability of unit's operation and the speed of load response are improved.