Reducing Energy Consumption in Smart Buildings Using Deep Learning and the LSTM Method

Abstract

The significant increase in energy consumption in the building sector has made its optimal management a priority. Accurate energy consumption forecasting is essential for designing intelligent control strategies, despite the complexities arising from environmental, physical, and behavioral factors. Drawing on data collected from smart buildings, this study develops and evaluates an energy consumption forecasting model based on a Long Short-Term Memory (LSTM) neural network. In this quantitative, data-driven study, time-series data comprising energy consumption, environmental variables, and temporal features were collected and preprocessed, after which a two-layer LSTM architecture was trained. The results demonstrated that the LSTM model achieved an accuracy of over 90% and significantly reduced forecasting errors compared with conventional methods and other machine-learning models. The model also showed a strong ability to identify complex temporal patterns and accurately track energy consumption trends. Moreover, the simulations confirmed that the model’s forecasts could facilitate load shifting, reducing peak demand by up to 15% and total energy consumption by up to 5%. Overall, this study confirms that the LSTM model is a powerful and efficient tool for forecasting and intelligently managing energy consumption in buildings, thereby contributing substantially to more sustainable management, improved energy efficiency, and reduced costs.