Research On Centrifugal Compressor Rapid Modeling Method Based On Data And Model Migration

Centrifugal compressor as an important part in Combined Cycle Power Plant(CCPP) gas system, its running status is directly related to stable and efficient operation of the whole system. In actual production, centrifugal compressor is easily influenced by the front steel or iron process. The inlet gas flow rate, pressure and temperature in compressor will have great fluctuation. The compressor often operates under variable operating conditions, and its output characteristics will have a certain drift. In order to avoid centrifugal compressor falling into surge or other unstable conditions, the optimize control is particularly important. However, the optimize control of compressor is based on accurate performance prediction. Therefore, it has important practical application value and theoretical research significance to study the rapid modeling method of compressor.In this thesis, we take centrifugal compressor in CCPP gas system as research object. Aiming at the problems of time-consuming and inefficiency in the existing modeling method, for the first time, we apply the idea of migration modeling to develop a new compressor model. The research contents include as following:(1) By analyzing the working principle, structure composition and physical characteristics of centrifugal compressor, using the energy and mass conservation law, a mechanism model of centrifugal compressor is established. Based on the mechanism model, two different specifications with exsting similar centrifugal compressor models are simulated. Meanwhile, the similarity of compression processes and the feasibility of model migration are also analyzed, which lay the foundation for migration modeling study of centrifugal compressor.(2) According to the family similarity of compression processes, a migration modeling strategy for new compressior which based on support vector mechine(SVM) is proposed. This strategy entirely extracts useful information from the existing compressor model, first of all the output of the existing compressor model is used as the additional input of new compressor model. Then, combining with a small amount of new compressor data samples to compensate for the differences between compression processes, and using SVM algorithm to quickly build a new compressor model. Finally, the predictive results of migrated model are compared with SVM model, the simulation results show that, in the case of a small amount of data samples, the migrated model has higher predictive accuracy and reliability.(3) Aiming at the limitation of migration modeling strategy for new compressor based on SVM, for example there is a certain subjectivity when collecting data samples of new compressor, and can not take into account the minimum number of experiment, etc. another migration modeling strategy based on Bayesian is used to establish a new compressor model. In order to obtain new migrated model, this strategy firstly adjusts Gaussian process model of the exisiting compressor by using scale-bias adjustment method. Then, based on the experimental data of new compressor using Bayesian to estimate parameters of new migrated model, and applying the sequential algorithm to iterative improvement of parameters estimation for new migrated model, until the parameters estimation satisfy the stopping rules. Finally, the simulation results show its feasibility.