A hybrid ANN-LSTM based model for indoor temperature prediction

To achieve the better control of heating, ventilation, and air conditioning (HVAC) with the energy saving target, the accurate indoor temperature prediction plays a fundamental role. However, existing temperature prediction approaches general overlook the relations with other enviromental variables, which results in unsatisfactory prediction results. Considering this defect, this paper developed a novel hybrid artificial neural network (ANN) and long short term memory (LSTM) model (HAL) to predict indoor multiple step ahead temperature. The proposed HAL can model the temperature pattern by LSTM and learn the relations with relative humidity and other related variables by ANN simultaneously. To avoid the error propagation, we make a direct map between the input feature and the predicted temprature. The proposed HAL is implemented on a realworld indoor environmental dataset and extensive experimental assessments are conducted. The results show that the proposed HAL outperforms LSTM, MLP and other data driven methods with strong robustness.