A Machine Learning Approach to Predict the Performance of Refrigerator and Air Conditioning Using Gaussian Process Regression and Support Vector Methods

Title:A Machine Learning Approach to Predict the Performance of Refrigerator and Air Conditioning Using Gaussian Process Regression and Support Vector Methods

Volume: 14 Issue: 4

Author(s): Harinarayan Sharma, Sonam Kumari, Aniket K. Dutt, Pawan Kumar*Mamookho E. Makhatha

Affiliation:

• Faculty of Engineering and the Built Environment, Department of Engineering Metallurgy, University of Johannesburg, John Orr Building, DFC, 25 Louisa St, Doornfontein, Johannesburg 2028,South Africa

Keywords: Coefficient of performance, Gaussian process, kernel functions, machine learning, nano-lubricants, refrigeration and air-conditioning, support vector.

Abstract:

Aim: The aim of this study is to develop machine learning models for the performance of refrigerator and air-conditioning system.

Background: The Coefficient Of Performance (COP) of Refrigerator and Air-Conditioning (RAC) is a complex function of evaporative temperature and concentration of nanoparticles in lubricants. In recent years, researchers have focused on experimental studies for the improvement of COP. Further, few researchers have applied simulation techniques such as fuzzy system, Artificial Neural Network (ANN), simulated annealing, etc. to Vapour Compression Refrigeration (VCR) cycle. Though, there is a scarcity of modeling research work for the performance of RAC system.

Objective: The study aims to develop the machine learning predictive models for the performance of refrigerator and air-conditioning system using experimental data.

Methods: The experiment was performed on VCR system to determine COP. Three different concentrations of lubricants (added 0.5, 1.0 and 1.5g nano-TiO₂ particle on 1 liter of Polyolester (POE) oil) were used. The experimentally calculated COP was used to train and test the machine learning models. Gaussian Process Regression (GPR) and Support Vector Regression (SVR) methods were applied to develop the models.

Results: The experimental result reveals that the COP increases with increasing the concentration (of nano particles) at a given temperature. The addition of 0.5 and 1.0g TiO₂ in the POE oil shows better rate of increment in the COP in comparison to addition of 1.5g TiO₂ in the POE oil. Machine learning models using GPR and SVR with RBF kernel function is the most appropriate machine learning model for the nonlinear relationship between the output parameter (COP) and the input parameter (evaporative temperature and concentration of TiO₂).

Conclusion: The present study was conducted to investigate the machine learning approaches for the performance of RAC system using experimental data sets. The experimental result shows that R134a and TiO₂-POE nano-lubricants work efficiently and the coefficient of performance of VCR system increases with concentration of nanoparticles. The developed model performance is compared using coefficient of correlation and RSME values. After comparison, it is concluded that RBF based GPR model is the best fit machine learning model to predict the COP in the context of any other model for this data set.

Cite this article as:

Sharma Harinarayan , Kumari Sonam, Dutt K. Aniket , Kumar Pawan*, Makhatha E. Mamookho , A Machine Learning Approach to Predict the Performance of Refrigerator and Air Conditioning Using Gaussian Process Regression and Support Vector Methods, Recent Patents on Mechanical Engineering 2021; 14 (4) . https://dx.doi.org/10.2174/2212797614666210125143810