Simulation and Prediction of Thermodynamic Performance of Reciprocating Compressor utilizing Physical Models Combining with Generalized Regression Neural Network

International Conference on Compressors and their Systems 2019

Qian, Lv^1 ; Xiaoling, Yu^1 ; Kun, Wang^1 ; Junchao, Ye^1 ; Shiyi, Fan^1 ; Wang, Xiaolin^1^2

School of Energy and Power Engineering, Xi'An Jiaotong University, Xi'an

710049, China^1

School of Engineering, University of Tasmania, TAS 7001, Hobart, Australia^2

关键词: Discharge pressures;    Discharge temperature;    Generalized Regression Neural Network(GRNN);    Generalized regression neural networks;    Polytropic exponent;    Prediction accuracy;    Suction temperature;    Thermodynamic performance;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/604/1/012023/pdf
DOI  :  10.1088/1757-899X/604/1/012023

来源: IOP

【 摘 要 】

Thermodynamic performance of a reciprocating compressor is generally evaluated or predicted by its physical models (PM). However in conventional PM, some key parameters are not easy to be determined and most time they can only be set values empirically. This article presented and experimentally verified physical models combining with generalized regression neural network (PMCGRNN) for the simulation and prediction of reciprocating compressor thermodynamic performance including the discharge temperature, and the volume flow rate. In PMCGRNN model of the compressor, firstly, the key parameters of compression polytropic exponent and the volume efficiency were obtained by generalized regression neural network (GRNN) with the input variables of suction temperature, suction pressure, discharge pressure, compressor rotate speed. Then the discharge temperature and the volume flow rate of the compressor were simulated respectively by their physical models (PM). The simulation results of discharge temperature and the volume flow rate by PMCGRNN were validated by a test bench of an air compressor. It was found that PMCGRNN has reliable prediction accuracy, and the relative errors of two PMCGRNN models are between -3.5% and +1.5%. and between -5% and 4%, respectively.