Digital twins model and its updating method for heating, ventilation and air conditioning system using broad learning system algorithm

Abstract

Digital Twins (DT) can be used for the energy efficiency management of entire life cycle of HVAC systems. The existing chiller models usually can not update in real-time, so they are not suitable for real-time interactions between DT models and real physical systems. In this paper, an intelligent DT framework is proposed for HVAC systems, which includes the equipment, data, simulation, and application layers. Broad learning system (BLS) is presented to build the simulation layer of the chiller and its DT platform. The basic BLS model is optimized to reach the best performance by choosing linear rectification function as activation function and setting batch size to 64 by enumeration method. The real HVAC system located in Zhejiang province is selected to verify the proposed method. For the first half year operation, the average mean absolute error, root mean square error and coefficient of determination (R2) of Multi-BLS model for nine chillers can reach 9.04, 15.20 and 0.98 respectively. For the second half year operation, the proposed method can be updated in 4.63s and its R2 is 0.95. Compared with conventional models, the proposed Multi-BLS model has better prediction precision and can be updated in real-time within a shorter time.