Research on machine learning based refrigeration system performance prediction

A combined experimental and machine learning approach was used to study the performance of the refrigeration system. Firstly, the thermal performance of the chiller refrigeration system under different operating conditions was evaluated experimentally and the results were analyzed; secondly, a support vector machine regression (SVR) prediction model was constructed using the experimental data to fit the mapping relationship between compressor speed, electronic valve opening, coolant inlet temperature and metering pump speed as input variables and system refrigeration capacity and COP as output variables. In addition, PSO was used to find the optimality of the kernel parameters ( g ) and penalty coefficients ( C ) in the SVR model, and the performance of the PSO-SVR model was evaluated. It has been found that increasing the compressor speed can significantly increase the refrigeration capacity, but also increases the energy consumption of the system; the optimized PSO-SVR model has good accuracy, both R 2 were improved compared to the SVR model, while the RMSE values decrease significantly by 51.75 % and 60.78 % respectively, which can be used as a new method to fit the non-linear relationship between the operating conditions and the performance parameters of the chiller refrigeration system.