A Distributed Fusion LSTM Model to Forecast Temperature and Relative Humidity in Smart Buildings

Abstract

The Heating, Ventilation, and Air Conditioning (HVAC) systems in the commercial buildings, consume near half of total building energy. Temperature and relative humidity as important indexes to evaluate the dynamic operation effect of HVAC systems, where their accurate forecasting is crucial to energy efficient management and indoor thermal comfort. However, the existing forecasting methods are suffering the poor prediction performance due to the strong correlation between temperature and relative humidity. To solve this problem, a novel distributed fusion long short-term memory (LSTM) network (DFL) is proposed, which utilizes the distributed data-fusion technology to achieve the prediction of temperature and humidity simultaneously. To obtain the optimal parameters setting, hyper-parameters analysis on the proposed DFL is conducted. A real-world building dataset is used to validate the potency, and the results show the DFL outperforms other state-of-the-art forecasting methods including SVR, fusion LSTM (FL) and classical LSTM (CL).