Chiller System Power Prediction by Physical-Informed Neural Network

In heating, ventilation, and air conditioning (HVAC) systems, chiller power model prediction is crucial for control and improving energy efficiency. However, in practical engineering scenarios, the chilled water supply temperature, chilled water flow rate, etc., are mostly set manually, and the system operates at a small number of fixed and sparse operating points. This leads to sparse data availability for model prediction, which seriously limits the prediction accuracy of data-driven models (such as fully connected neural networks). To overcome the above problems, this paper introduces a Physics-Informed Neural Network (PINN), and by embedding physical knowledge, performs predictive modeling of the power of the core equipment in the chiller system—the chiller, condenser water pump, and chilled water pump. Based on the real industrial data of a large building in southern China, the qualitative and quantitative verification shows that the model proposed in this paper has significant advantages in prediction accuracy compared with traditional methods.