A deep learning-based energy consumption prediction model based on occupants and space appliance in mixed-mode residences

Abstract

The behavior of the resident plays an indispensable role in the prediction of energy consumption by the buildings. But, the unreliability of the activities of the occupants makes the prediction of the energy consumption model complicated. The occupant behavior, characteristics of the building and the indoor or outdoor climate highly impact the performance of buildings in terms of energy utilization. The simulation tools used for energy consumption prediction do not provide precise results. They give outputs based on the assumptions, not by computing the energy usage of the buildings. So, the prediction results are not accurate. This work introduced a new prediction model using deep learning strategy with parameter optimization. Initially, data related to energy consumption is attained manually from buildings. The occupants, space and appliances are the inputs obtained from the data, and it is transferred to the energy consumption prediction stage. Here, the prediction is performed by Multi Serial Cascaded Deep Networks (MSCDN) with attention mechanism, where the deep learning structures like Autoencoder, Deep Temporal Convolution Network (DTCN), and Long Short-Term Memory Networks (LSTM) are utilized to form the MSCDN. Here, the parameters present in the MSCDN are optimized by Hybridized Archimedes and Transit Search Optimization (HATSO). The developed prediction model has been improved by the use of parameter optimization.