Evaluation of machine learning-based applications in forecasting the performance of single effect absorption chiller network

The present study aims to predict the coefficient of performance and thermal energy consumption of an absorption chiller network, using three widely-used machine learning methods of the artificial neural network, support vector machine, and genetic programming. To this aim, a case study was conducted on the Marun petrochemical company in Iran. The genetic programming was used to estimate new formulas for the functions in terms of operational variables. Then, using the optimization algorithm, the optimal load of each chiller in the network was obtained. The results revealed that the artificial neural network technique has the highest prediction accuracy among the mentioned methods, in which the mean square errors of the performance coefficient and thermal energy consumption of chiller are 1.683 x 10(-8) and 8.157 x 10(-8), respectively. Also, for the support vector machine and genetic programming methods mean square errors are 1.627 x 10(-3), 1.135 x 10(-3) and 2.187 x 10(-3), 4.358 x 10(-3), respectively. The new estimated formulas for the performance coefficient and thermal energy consumption of each chiller based on the genetic programming have acceptable accuracy and their coefficients of determination are 0.97093 and 0.95768, respectively. Moreover, given the constant operating variables, if the cooling load of each chiller in the network is optimally selected, the thermal energy consumption of the network will decrease averagely by 2.1 % and the performance coefficient of the network will increase by 1.3 %.