Abstract: In order to improve the accuracy of thermodynamic modeling for centrifugal compressors, an adaptive modeling method for thermodynamic model and characteristic curve of centrifugal compressors based on field test data was proposed.During the modeling, the field installation errors and the degeneration during operation were considered, and the accuracy of thermodynamic calculation was improved by the modified similarity conversion method. For the inaccurate general characteristic curve, a neural network prediction model based on multi-layer perception was established to generate new characteristic curve, and the characteristics was adjusted with the particle swarm optimization algorithm, so that the output at the design points and the off-design points under variable working conditions met the matching accuracy with the field test data. With this adaptive method, a thermodynamic model was established for PCL803 centrifugal compressors from Italy GE Company. The results show that the calculation error of the design points and the off-design points under variable working conditions after performance adaptation is reduced significantly compared with that before adaptation when the general characteristic curve is used. The research results could provide basis for the condition monitoring, performance prediction and fault diagnosis of centrifugal compressors.