Application of machine learning techniques in operating parameters prediction of Stirling cryocooler

The Stirling cryocoolers are widely used in the military and the aerospace fields due to many advantages such as high efficiency and compact structure. The output performance is affected by three parameters: compressor stroke, expander stroke, and the phase shift between them. How to quickly and effectively adjust the three parameters to meet the required cooling capacity with a higher COP is of great significance to the actual engineering application. When the cooling demand is changed, it takes a long time to calculate the corresponding high-efficiency operating parameters using the Stirling cryocooler simulation model on the market. In this paper, the prediction models of the optimal combination of operating parameters based on particle swarm optimization and commonly used machine learning techniques of back propagation neural network, support vector regression and random forest regression are established for the Stirling cryocooler. Besides, the impact of the two data preprocessing methods (min–max normalization and principal component analysis) on the PV power models and compressor stroke regression models based on three different machine learning techniques is analyzed, which shows that the use of principal component analysis can significantly enhance the performance of the back propagation neural network, and the normalization of the cryocooler parameters can improve the convergence of the support vector regression and the random forest regression. The results show that the regression models established by the three machine learning algorithms all have high accuracy and generalization. The regression model based on support vector machine has the best performance on the prediction of PV power and compressor stroke with small mean squared error, mean absolute error, average relative error, and high Pearson correlation coefficient. In addition, the research on the prediction models for the output performance is beneficial to propose a new control strategy, so as to improve the control system of the Stirling cryocooler.