A Digital Twin Model for COP Prediction in Refrigeration System using Combined Machine Learning Method

Abstract

Accurate prediction of coefficient of performance (COP) in refrigerant system is beneficial to the overall system performance evaluation and dynamic operation optimization. In this paper, a digital twin model of COP prediction in refrigeration system based on combined machine learning method is proposed. Firstly, several influencing factors are selected guided by the mechanism knowledge, and the correlation coefficients between impact factors and COP value are calculated to further verify their importance in COP prediction. Secondly, the convolutional neural network (CNN) and multilayer perceptron (MLP) with intrinsic weight optimization are used to capture the characteristics between the influencing factors and COP, while the long short term memory (LSTM) method is used to capture the characteristics of the time series data. Meanwhile, the weights of the CNN, LSTM and MLP methods are optimized in real time by the particle swarm optimization (PSO) algorithm to minimize the difference between the COP value of physical system and the predicted COP value of digital twin model to realize online self-update. Finally, simulation results show that the MAE, MAPE and RMSE obtained by the proposed method are respectively 7.01 %, 1.71% and 0.08% lower than those obtained by CNN method, 0.92%, 0.24% and 0.01% lower than those obtained by LSTM method, and 2.41%, 0.58% and 0.02% lower than those obtained by MLP method, which have proved that the prediction result of proposed digital twin model for COP prediction is more accurate.