Transfer Learning Predicts VRF Unit Noise Using Thermodynamic and Vibration Signals

A novel unsupervised transfer learning approach, named Domain-invariant Partial Least Squares (Di-PLS), has been introduced by researchers to forecast the second-order harmonic (2f) noise in variable refrigerant flow (VRF) outdoor units. This 2f noise, produced by twin-rotary compressors, represents a major low-frequency noise source that varies with environmental thermal loads and valve openings, complicating its evaluation through standard models. The investigation develops prediction models utilizing thermodynamic and acceleration data, contrasting Di-PLS with conventional Partial Least Squares (PLS). Findings indicate that Di-PLS significantly surpasses PLS by effectively extracting common features across conditions and reducing distribution differences between source and target domains. The study can be accessed on arXiv with ID 2605.00895.