Research On Modeling Method Of Centrifugal Compressor Based On Multi-model Migration Strategy

As the core energy-conversion equipment in modern industry energy conversion,centrifugal compressor is widely used in many fields,such as petroleum and petrochemical,metallurgical mining,ventilation and refrigeration.Its safe,stable and economical operation have always been the goal pursued by modern industry.However,in actual production,the operation state of the centrifugal compressor is highly susceptible to fluctuations in working conditions,fluid medium,etc.,and optimization and control of its operation state is particularly important.The basis of the operation state optimization control is the accurate and reliable performance prediction model.Therefore,it is of great significance to research the rapid establishment of an accurate and reliable performance prediction model for centrifugal compressor.This paper employs the large-scale high-energy-consumption centrifugal compressors in the complex industrial process as the study subject.Due to the short operation time and lack of modeling data,it is difficult to develop an accurate performance prediction model for the new centrifugal compressor.Based on the idea of developing the new process model by migrating the information of similar processes,this paper studies the migration modeling method of centrifugal compressor by the means of model migration,Bayesian model averaging and multi-task least squares support vector machine,and so on.Then,this paper proposes a lower cost modeling method of centrifugal compressor based on multi-model migration and a joint modeling method for performance prediction model of multi-centrifugal compressor based on multi-task learning,respectively.The main research contents are as follows,(1)Principle and characteristic analysis of centrifugal compressor.Based on the previous research results,this paper analyzes the operation mechanism of centrifugal compressors and establishes a mechanism model for predicting the performance of centrifugal compressor.Then,the mechanism model is employed to study the factors that influence the performance of centrifugal compressor,which is the foundation of the verification for the subsequent chapters.(2)A lower cost modeling method of centrifugal compressor based on multi-model migration.Model migration between two similar centrifugal compressors does not fully exploit the potentially useful modeling information in the similar centrifugal compressors,which results in high modeling costs and long cycle times.For this problem,this chapter proposes a lower cost modeling method of centrifugal compressor based on multi-model migration strategy.This method can fully exploits the useful information in multiple similar centrifugal compressor models through the multi-model migration strategy,which helps the new compressor to establish an accurate performance prediction model quickly.In addition,the method uses the Bayesian model averaging,nested Latin Hypercube Design and other algorithms to ensure that the data required for modeling is minimal,which reduce the cost of modeling further.Finally,the effectiveness of the proposed method is verified by simulation experiments.(3)A joint modeling method of multi-centrifugal compressor for performance prediction model based on multi-task learning.This chapter proposes a Joint modeling method to realize the modeling of multi-centrifugal compressors simultaneously.This method views the centrifugal compressor model as two parts: common features and unique features.According to the correlation of similar compressor modeling tasks,under the multi-task learning mechanism,this method uses the information of the similar compressors to learn the common features,and employs the data of target compressor to learn the unique features simultaneously,and then develop the models of multiple centrifugal compressors.The essence of the method is to use the modeling information from similar centrifugal compressors as the bias signal to improve the accuracy of the models of multiple centrifugal compressors.Simulation experiments show that the proposed method can realize modeling of multiple similar centrifugal compressors that improve modeling efficiency and reduce modeling costs.