Application of BP Neural Network for Pre-Dehumidification Time Prediction of Capillary Ceiling Radiant Cooling Panel Air Conditioning System

Abstract

Pre-dehumidifying the room is generally needed before the capillary ceiling radiant cooling panel (CCRCP) air condition system is turned on. Accurate pre- dehumidification time is critical for condensation prevention and energy usage. The pre- dehumidification time, which is related to multiple variables with complicated correlation relationship, is difficult to be calculated by conventional methods. Therefore, BP neural network is considered to be applied to predict the pre-dehumidification time. In this study, a dynamic model of CCRCP + displacement ventilation air conditioning system was built to simulate the pre-dehumidification process in TRANSYS. And then BP neural network was established, it takes the indoor and outdoor temperature and humidity conditions at 7:00 in every morning as the influencing factors and predict the optimal pre-dehumidification time for each day. The results show that the mean square error (MSE) of the BP neural network training process is 1.90958×10−4, the correlation coefficient R between the training data and the sample data reaches 0.99906, and the correlation coefficient R between the predicted data and the sample data reaches 0.99897. The BP neural network can reflect the intrinsic relationship between optimal pre-dehumidification time and input variables, and has high accuracy in predicting optimal pre-dehumidification time of CCRCP air conditioning system.