Refrigeration Performance Analysis Using Artificial Neural Network

Abstract

AbstractThis report aims to introduce a very unique and innovative approach for the performance analysis of a Vapor Compression Refrigeration System (VCRS) which is applicable for systems using any of the refrigerants available with known properties. The use of Artificial Neural Network (ANN) has been proposed to determine the performance via Coefficient of Performance (COP) of the VCRS System. The ANN prediction model is developed in R programming using the H20 library. For training the neural network, the data is obtained by using Engineering Equation Solver (EES) software which was simulated for 50 different values of input parameters to obtain the Performance (COP) of the VCR cycle for 15 different refrigerants. In this ANN model, the input layer consists of Condenser temperature (T_C), Evaporator temperature (T_H) and different properties of the refrigerants. After training the data multiple times and simultaneously checking the accuracy, 100 hidden layers and 80 neurons in each hidden layer are finally used which provide the best accuracy. The prediction observed from the ANN model is within the acceptable margin of error. The maximum error obtained is less than 5%. The RMSE and Mean Residual deviance values are also within acceptable limits. The EES software used here is just a substitute for the real experimental setup which this approach can also be used to analyze so that time and resources required for a traditional approach can be saved. The ANN algorithm is further utilized to predict optimum conditions which provide the maximum performance for each input parameter by running continuous simulations on them and thereby representing them in the form of charts for better understanding.