Applicability of energy consumption prediction models in a department store: A case study

Obtaining an accurate picture of energy consumption in public spaces is a critical first step for predicting and enhancing energy efficiency. In practice, such calculations are often complicated, necessitating use of artificial intelligence models which can account for multiple factors. In this study, we obtained four years of daily electricity usage, people flow, and ambient factor data from a department store in Taiwan to build an artificial intelligence model for predicting energy consumption. Previous literature on public spaces have mainly used Long Short-Term Memory (LSTM), gated recurrent unit (GRU), and other recurrent neural network (RNN) mechanisms. However, as our dataset spanned four years, we hypothesized that seasonal and trend decomposition using Loess (STL) was applicable as our data likely exhibited long cyclic patterns. We built and compared prediction results of three neural networks: LSTM-classic, STL, and GRU. As expected, the STL model provided better overall results (indicating a strong seasonality component in our data), as well as an optimal balance between accuracy and processing time. This case study provides a framework for others seeking to build neural networks for predicting and managing energy consumption in public spaces, and forms a theoretical basis for further discussion of energy efficiency topics.