Predictive Modeling of Energy Consumption for Building Automation Systems Using Deep Learning

In the realm of building automation systems, efficient energy consumption modeling stands as a crucial element for sustainable and intelligent environments. This paper presents a predictive modeling approach utilizing deep learning networks like Long Short-Term Memory (LSTM) neural networks, alongside comparisons with Gated Recurrent Unit (GRU) and Bidirectional LSTM (Bi-LSTM) architectures, for predicting energy utilization within buildings. The study employs a smart home dataset from the UMASS Repository, demonstrating the model's effectiveness in capturing temporal dependencies and delivering accurate predictions. The methodology encompasses comprehensive data preprocessing, model training, and evaluation using key performance metrics. Significantly, the obtained results showcase the superior performance of the LSTM model over GRU and Bi-LSTM, with a Mean Absolute Error (MAE) of 0.2662, Root Mean Squared Error (RMSE) of 0.4067, and Mean Squared Error (MSE) of 0.1654. These metrics underscore the model's capability to provide reliable and accurate predictions, thereby contributing to optimizing energy management within building automation systems. The demonstrated success of the predictive model highlights its potential significance in advancing sustainable practices, promoting resource efficiency, and facilitating intelligent decision-making within the realm of intelligent buildings.