Development of a method for predicting the transient behavior of an absorption chiller using artificial intelligence methods

Abstract

Absorption chillers are complex equipment compared to mechanical chillers due to their nature of heat and mass transfer processes, working fluids used, and their limitations, such as crystallization problems and dealing with vacuum pressures. Even more, when their dynamic behavior is conducted, it leads to a challenging study due to the simultaneous heat and mass transfer present in the thermal compressor. In line with this, the need to know or measure the values of internal parameters, such as temperature, concentration, and internal flows. To perform the measuring leads to using complex physical models to simulate the transient behavior, making it hard to adapt more efficient control strategies. Hence, this work proposes a methodological strategy to determine the transient behavior of a LiBr/H2O single-effect absorption chiller through intelligent regression methods. The methodology was developed by analyzing external data from an absorption refrigeration system's hot, cold, and chilled water circuits, characterizing regression models exclusively with temperature inputs, and using four machine learning methods, such as linear regression, decision tree, random forest, and artificial neural network. One novelty of the work is the integration of the first law of thermodynamics and the heat exchanger characteristic equation with machine learning methods, such as linear regression (LR), decision tree (DT), random forest (RF), and artificial neural network (NN) to estimate the dynamic of absorption chillers through exclusively external data such as temperature and flow, and nominal data of the heat exchangers. Another novelty is the capacity to determine the internal conditions of the equipment, considering a full or partial operation. The validation and calibration procedure established that all the methods applied (LR, DT, RF, and NN) presented a good fit showing values above 0.95 of the correlation coefficient R2 and RMSE values less than 0.1. The results showed that the strategy allows for estimating the transient behavior of the outlet temperature profiles of the hot, cold, and chilled water circuits. The behavior parameters were calculated with reasonable accuracy showing errors of less than 1 % for all methods applied. The thermal COP value estimated was approximately 0.71 ± 0.05, within the values presented by the equipment manufacturer. Regarding the results, the proposed strategy represents a robust and accurate dynamic analysis tool for absorption chillers operating at full or partial load.