Multi-input multi-output (MIMO) artificial neural network (ANN) models applied to economized ccroll compressors

摘要

Predicting the compressor performance is an essential part in the design and optimization ofHVAC&R equipment. The AHRI standard polynomial equations are well suited to map theperformance of single-stage fixed-speed positive displacement machines, but present a number ofwell documented shortcomings especially in the case of compression enhancements (e.g. oilflooding,injection, variable-speed). To this end, a Multi-Input Multi-Output Artificial Neural Network(ANN) model has been developed to map the performance of single-phase and two-phase injectedscroll compressors. The ANN models are based on a multi-layer structure whose number ofneurons has been optimized to obtain high-accuracy predictions. The models have beendeveloped by using the open-source Keras package. The trained ANN models have beencompared with the current state-of-the-art correlations available and they outperformed the existingcorrelations in terms of accuracy.