Deep-learning neural network prediction of a solar-based absorption chiller cooling system performance using waste heat

Abstract

In this research, the neural network was used to convert a dual-effect absorption chiller that is housed in the pricey TESS library of the TRNSYS software into a function in MATLAB software. Real catalog data from a prominent manufacturer are used to create this absorption chiller. By considering characteristics including the temperature of the cold input and output, the temperature of the cooling water input, and the coefficient of nominal efficiency of the equipment, it was concluded that the neural network would be used to compute the real operational coefficient of the equipment after performing research on the inputs and outputs of the absorption chiller that were found to be the most effective. The deep learning neural network utilized in this research was subjected to numerous investigations. In addition, the influence of concealed layer neurons, stimulation function of concealed layer neurons, and dropout probability on network precision were studied. According to the findings of the research, the most accurate network is one that has three concealed layers, each of which contains ten neurons, a hyperbolic tangent activation function, and a dropout probability of 40 %. It can achieve a correlation coefficient of around one between anticipated and real data. MATLAB software was used to conduct a parametric evaluation system on this dual-effect absorption chiller after the neural network was constructed. The findings indicated that as the determined temperature of the cold water output and the nominal efficiency coefficient of the system enhanced, the quantity of energy produced by the generator and the temperature of the output cooling water reduced.