Research On Modeling Theory And Method For 1000MW Ultra Supercritical Unit

The primary task of improving the control performance of 1000 MW Ultra Supercritical (USC) once through unit is to understand its operation characteristic and then establish appropriate mathematicl models which is suitable for controller design. Based big historical data, identification method of local linear model and global nonlinear model for multivariable system is presented in this dissertation. Main contents are given as follows:(1) The background and significance on modeling theory and method for 1000MW USC unit are introduced, current research status of thermal power unit modeling, multivariable system modeling and modeling based on big historical data are reviewed.(2) Common mathematical model structures are analyzed and both the advantages and disadvantages of each model structure are summarized. Research on the characteristics of mechanism analysis method and experimental identification method shows that it is much difficult to model a complex system using a single modeling method. Appropriate use of hybrid modeling method is a good solution. Suitable choice of mathematical model and modeling method depends on many factors, i.e., system status, modeling objectives, modeling accuracy, modeling cycle and cost requirements. Besides, the researcher’s experience is also a critical factor in modeling. It is important to choose appropriate model and method according to practical situation.(3) Based on the quantum computation theory and PSO algorithm, this dissertation presents a novel double quantum particle swarm optimization(DQPSO) algorithm in which the diversity of population particle is increased by simultaneously quantizing both the coding and evolutionary search strategy. At the same time, elimination of velocity vector results in a evolution equation with simpler structure and less experience parameters so that it is easier to control. Several benchmark functions are used to validate the optimizaton capability of the proposed DQPSO algorithm. In order to improve the optimization efficiency of the algorithm further, two improved DQPSO algorithms are proposed. The first one is DQPSO algorithm with an additional acceleration factor (A-DQPSO), another is DQPSO algorithm based on the Memetic framework (M-DQPSO). Test results show that the DQPSO algorithm is suitable for general high dimensional (below 50 dimensions) function optimization, A-DQPSO is suitable for higher dimensional (below 100 dimensions) function optimization and M-DQPSO is suitable for ultra high dimensional (more than 100 dimensions) function optimization.(4) The characteristics of big historical data is analyzed, and the methods of data sample selection and pretreatment are designed for thermal system modeling. Based on big historical data, a hybrid modeling method combining mechanism modeling with intelligent modeling for multivariate system modeling is proposed. Firstly, the model structure and the initial range of the parameters are determined by setp response experiments on the mechanism simulation model. The accurate transfer function model is then obtained through the use of intelligent parallel algorithm and historical operating data of a power plant. The impact of large-scale and frequent step response experiments on the production site can be avoided in this method. At the same time, the initial model structure and the parameter range, which are difficult to determine in multi-variable intelligent optimization, can also be ascertained. Finnaly, this new identification idea is used to identify the 1000MW USC unit coordinated control system based on selected historical data and DQPSO algorithm. The system model obtained can be used for controller design and optimization.(5) A linear parameter time-vary ing (LPV) model reprensention based on unit load is proposed. The LPV model established by A-DQPSO algorithm based on historical data can descirbe the global dynamic properties of the superheated steam temperature system and can be used for control study. Identification results show that the LPV model is suitable for thermal system modeling. As for the model identification of multivariable thermal system, the LPV model with networked internal structure can be used to represent the global dynamic properties by making use of time-varying characteristics of multiple standard feature variables. This networked LPV model can precisely describe the nonlinearity of the system, and has a clear physical interpretation. Finnaly, a global networked LPV model of 1000MW USC unit coordinated control system is obtained based on M-DQPSO algorithm and big historical data.