Development and validation of intelligent load control for VRF air-conditioning system with deep learning based load forecasting

Abstract

Variable refrigerant flow (VRF) air-conditioning systems have seen significant growth in Asia and its application is expanding globally. Despite the expanded application, most previous studies have focused on developing fault detection and diagnostics to achieve energy-efficient operations than on predicting power consumption. It's very difficult to predict the electrical consumption owing to its complex system configuration and various control strategies. A new control strategy is described for optimal adjustment of the desired target level based on time series forecasting using the optimized sequence to sequence model for VRF systems. Sequence to sequence (seq2seq) model with attention mechanism and Bayesian optimization is developed to predict accurate hourly and daily forecasts and rapid feedback control for fluctuating power consumption for VRF systems. The optimized seq2seq model is integrated into the intelligent load control (ILC). ILC can be used to manage VRF systems by dynamically prioritizing indoor units for curtailment using both quantitative inputs and qualitative rules. Overall, the results demonstrate that the deep learning based control allows coordination of the controllable loads of VRF systems in three commercial buildings. ILC with deep learning manages the power consumption within a desired target level, as well as indoor temperature reflecting the status of controlled indoor units, as objective functions of the control algorithm.