Multi-step prediction of electronic expansion valve opening of a low-carbon air-conditioning system with condensation heat-recovery unit

Condensation heat recovery provides an energy-saving approach for air reheating in air-conditioning systems, but its dynamic performance depends strongly on the regulation of refrigerant distribution. The electronic expansion valve (EEV) controls the refrigerant flow entering the heat-recovery branch, and accurate short-horizon prediction of EEV opening can support predictive operation and supply-air state regulation. In this study, a VMD-GRU-Attention framework was proposed for direct multi-step prediction of EEV opening in an air-conditioning system with a condensation heat-recovery unit. Variational mode decomposition (VMD) was first used to decompose the original EEV opening sequence into modal components with different frequency characteristics. A gated recurrent unit (GRU) network was then used to extract temporal dependencies, and an attention mechanism was introduced to enhance the contribution of informative historical states. Experimental data collected from an actual system were used for model evaluation. Compared with RNN, TCN, LSTM and GRU, the proposed model achieved lower prediction errors and a more concentrated absolute percentage error distribution. At the 10th prediction step, MSE, RMSE, MAE and MAPE were reduced by 57.78%, 35.02%, 35.75% and 35.78%, respectively, compared with GRU. The results demonstrate the framework's effectiveness for EEV opening prediction and provide a basis for future control-oriented studies.