A low data requirement model of a variable-speed vapour compression refrigeration system based on neural networks

In this work a model of a vapour compression refrigeration system with a variable-speed compressor, based on a black-box modelling technique, is presented. The kernel of the model consists of a full customized radial basis function network, which has been developed to accurately predict the performance of the system with low cost data requirement in terms of input variables and training data. The work also presents a steady state validation of the model inside and outside the training data set, finding, in both cases, a good agreement between experimental values and those predicted by the model. These results constitute a first step to go through future research on fault detection and energy optimisation in variable-speed refrigeration systems.