Compressor performance modelling method based on support vector machine nonlinear regression algorithm.

Area of Science:

• Mechanical Engineering
• Computational Fluid Dynamics
• Machine Learning

Background:

• Compressor performance modeling is crucial for predicting thermodynamic behavior under varying operational demands.
• Incomplete compressor characteristic maps, lacking on-design operating points, pose challenges for accurate performance calculations.
• Existing neural network models (BP, RBF, Elman) have limitations in interpolation and extrapolation accuracy.

Purpose of the Study:

• To develop a novel compressor performance modeling method using support vector machine (SVM) nonlinear regression.
• To evaluate the proposed SVM method's accuracy and computational efficiency against established neural network algorithms.
• To demonstrate the SVM method's capability in handling incomplete compressor characteristic maps.

Main Methods:

• Implementation of a support vector machine nonlinear regression algorithm for compressor performance modeling.
• Comparative analysis with Back Propagation (BP), Radial Basis Function (RBF), and Elman neural networks.
• Evaluation based on interpolation/extrapolation accuracy and calculation time.

Main Results:

• The SVM method demonstrated superior interpolation and extrapolation performance compared to BP, RBF, and Elman networks.
• For flow characteristic maps, the SVM method achieved a total RMSE of 2.72%, outperforming the Elman algorithm by 47%.
• For efficiency characteristic maps, the SVM method achieved a total RMSE of 1.81%, outperforming the BP algorithm by 35%.

Conclusions:

• The proposed SVM-based compressor performance modeling method is effective for handling incomplete characteristic maps.
• The SVM method offers improved accuracy and real-time performance over traditional neural network approaches.
• This approach enhances the ability to accurately calculate compressor thermodynamic performance under variable working conditions.