Neural-Network-Based Polynomial Correlation of Single- and Variable-Speed Compressor Performance

The compressor is one of the major components in a vapor-compression refrigeration system. A neural-network-based polynomial correlation method of positive-displacement compressor performance has been developed that can be applied to both single-speed and variable-speed compressor families. The multi-layer perceptron neural network was used as a universal function approximator. To align with and extend the ARI ten-coefficient correlation method (ARI 1999), the third-order polynomial transfer function is customized in the hidden layer and the pure linear function is adopted in the output layer of the neural network. The ARI ten-coefficient correlation has been proven as a special case of the proposed neural network. The new neural network method can be easily extended to multi-input/multi-output cases. In particular, in modeling the performance of a single-speed or variable-speed compressor family, this method gives less than 1% standard deviations and ±3% maximum deviations against manufacturer data.