An online self-learning modeling approach for absorption refrigeration systems

Given the dependence of the absorption refrigeration system on the heat source and the complexity of the operating conditions, it is very difficult to accurately predict the cooling capacity of the system when the heat supply from the heat source is unstable. In this paper, an online self-learning modeling method for predicting the cooling capacity of the absorption refrigeration system is proposed using neural networks. Initially, the input-output structure of the model is determined by examining the factors that impact the cooling capacity of the absorption refrigeration system. Subsequently, the internal structural framework of the model is developed by seamlessly combining the fully connected neural network with the residual neural network. Finally, the model parameters are determined by applying the gradient descent method to experimental data. For the implementation of the online self-learning for models, the baseline model, which is employed for the primary screening of the online data, is first trained by the small-sample offline dataset. Subsequently, the correction model, which is utilized for the secondary screening of online data and the online refinement of model parameters, is trained using the online data. The experimental results illustrate that the predictive accuracy of the developed model shows incremental improvements of 0.34 %, 0.52 %, and 0.86 % over three rounds of online self-learning in comparison to the baseline model. This finding validates the capability of the proposed system model to enhance its predictive accuracy and broaden its adaptive range consistently through the online self-learning feature.