Awareness-guided Incremental Control Optimization for Chilled Water System with Deep Learning Model under Cold-Start Scenarios
Xinbin Liang,Zheming Zhang,Fei Zheng,Siliang Chen,Xinqiao Jin,Zhimin Du
DOI: https://doi.org/10.1016/j.buildenv.2024.112092
IF: 7.093
2024-01-01
Building and Environment
Abstract:For heating, ventilation, and air conditioning (HVAC) systems, the technology of optimal control can achieve significant energy savings by resetting the control setpoint in responding to the change of user load and weather condition. However, traditional method will fail in cold-start scenarios, in which the diversity of history data is limited. Under this case, significant prediction error will occur in extrapolation space, and the optimized control strategy cannot fully achieve energy saving potential or even increase energy consumption. To solve this problem, here we propose a novel awareness-guided incremental control optimization method. Its basic idea is to explicitly consider the prediction error of energy model during the control optimization process. The deep ensemble algorithm is firstly adopted to capture the prediction error. And the optimization process will be aware of such potential error, and make a tradeoff between conservative state and exploratory state. The control optimization process will start with the known optimal setpoint (conservative state) to achieve stable energy saving, and then gradually explores other unknown candidates (exploratory state) to achieve further energy saving. To validate our method, a virtual test bed of chilled water system is developed by Modelica language to compare fixed setpoint strategy, traditional method, and proposed method. The results demonstrate that the proposed method can avoid negative energy saving ratio during the early stage of control optimization. And it can also find optimal control strategy faster. Comparing with traditional method, the awareness-guided method can achieve further 2.9 % energy saving ratio, and the overall energy saving is between 6.1% and 13.9 %.