Deep learning-based defrosting control strategy for air-source heat pumps utilizing electronic expansion valve opening patterns

초록

Frost accumulation in outdoor heat exchangers significantly degrades the heating performance of air-source heat pumps under frosting conditions. Conventional time-temperature control (TTC) methods often lead to maldefrosting, characterized by premature or delayed defrosting initiation, because they rely on fixed time thresholds that fail to adapt to dynamic environmental conditions. Although model-based approaches can mitigate this issue, they typically require additional internal sensors. To address these limitations, this study proposes an electronic expansion valve (EEV)-based defrosting control (EDC) strategy that predicts the heating capacity degradation ratio (HDR) using the EEV opening decreasing ratio (EDR) and ambient data. The originality of this study lies in utilizing EEV opening patterns to identify frosting-induced heating performance degradation without additional internal sensors. Four deep learning architectures-fully connected deep neural networks (FCDNN), convolutional neural networks (CNN), long short-term memory (LSTM), and gated recurrent units (GRU)-were developed to learn the nonlinear relationship between EDR and HDR. Among these models, the CNN model showed the highest overall HDR prediction accuracy on the test data, with an R2 of 0.973 and a MAPE of 1.9%. Based on the experimental data under various frosting conditions, the CNN-based EDC strategy predicted the defrosting initiation time within 20 s of the optimal points and improved the duration efficiency performance index by up to 7.2% compared with the conventional TTC method. These results demonstrate the feasibility of using EEV opening patterns for defrosting control without additional internal sensors; however, further validation using different ASHP configurations and field-operation data is required before broader applications.

키워드