LSTM-Assisted Heating Energy Demand Management in Residential Buildings

Smart heating system is one of the most efficient ways to realize indoor heating comfort. As the most energy consumption in the residential buildings is related to the heating, introducing an efficient energy management algorithm is so important for heating operation management. Accordingly, this chapter focuses on residential buildings’ heat load prediction using available historical and affective parameters in structure of building. To do this, a deep learning technique called long short-term memory (LSTM) is proposed for the heating load prediction of residential buildings. LSTM, also known as recurrent neural networks or applications of deep learning networks, has many applications such as data mining, prediction, system modeling, optimization, pattern recognition, etc. LSTM finds interrelations of input data, which cause to select optimal structures for models. The dataset used in this chapter consists of 768 samples, and each sample has 8 features which produced by using 12 different building shapes in Ecotect software. There is a heating value (as target) for each of these samples. The proposed method is taught using training data pertaining to the characteristics of each sample. Trained networks are saved as a black box containing the features extracted from each sample. This process is followed by forecasting heating load for unknown data or new conditions using these saved networks.