Abstract: Due to the wide range of cooling load of variable refrigerant volume, VRV system, the existing constant setting point of evaporation temperature control strategy always causes issues such as low performance efficiency and unsteady room temperature and humidity, especially with low part load ratio. A new kind of evaporation temperature control strategy of VRV system has been developed to improve the energy efficiency of the system and indoor climate control by setting variable evaporation temperature with predicted load. Thus, the key of the control strategy is to predict load on-line rapidly. Two kinds of data-driven models (Elman and along short-term memory neural network models, LSTM) have been developed to replacing the physical model, which is too complicated and time-consuming for VRV control module. The mean relative deviations of Elman and LSTM models are 8.6% and 2.0%. And LSTM model has been selected to couple with the control module due to its rapid computation speed and high precise. Furthermore, the effect of the load-prediction based variable evaporating temperature control strategy has been verified by the co-simulation of a VRV system model and a building model. According the simulation results, the VRV system applied the load-based variable evaporation temperature control strategy could consume less power than that of constant evaporation temperature control strategy under the low-load condition. And about 14% energy saving could be achieved by the new control strategy during the cooling operation period in Guangzhou, while the requirement of indoor thermal comfort has been satisfied. The load-based control strategy could be applied in remote intelligent cloud control system for VRV to reduce building energy consumption and carbon emission.

Key words: Load prediction, Variable refrigerant volume system (VRV system), Energy-saving control