Artificial intelligence models for refrigeration, air conditioning and heat pump systems

Artificial intelligence (AI) models for refrigeration, heat pumps, and air conditioners have emerged in recent decades. The universal approximation accuracy and prediction performances of various AI structures like feedforward neural networks, radial basis function neural networks, adaptive neuro-fuzzy inference and recurrent neural networks are encouraging interest. This review discusses existing topographies of neural network models for RHVAC system modelling, energy prediction and fault(s), and detection and diagnosis. Studies show that AI structures require standardization and improvement for tuning hyperparameters (like weight, bias, activation functions, number of hidden layers and neurons). The selection of activation functions, validation, and learning algorithms depends on author’s suitability for a particular application. Backpropagation, error trial selection of the number of hidden layer, and hidden layers’ neurons, and Levenberg–Marquardt learning algorithms, remain prevalent methodologies for developing AI structures. The major limitations to the application of AI models in RHVAC systems include exploding or/and vanishing gradients, interpretability, and accuracy trade off, and training saturation and limited sensitivity. This review aims to give up-to-date applications of different AI architectures in RHVAC systems and to identify the associated limitations and prospects.