Smart Prediction Models for Energy Efficiency in Smart Residential Buildings: A Deep Learning Approach

Using power in houses has given rise to several ecological issues, including air pollution. Predicting building energy use is essential for better regulating or reducing energy usage. Machine Learning (ML) techniques have been used to estimate the power use of operating buildings. However, there has been little research on using ML algorithms to anticipate the prospective energy usage of buildings during the initial design phase. When lowering energy-inefficient buildings, it is crucial to tackle the underlying issue: accurately anticipating energy consumption before construction. This is necessary to prevent the future development of environmentally detrimental buildings. This research introduces a new and innovative Hybrid Deep Learning Model for Predicting Energy Efficiency in Smart Residential Buildings (HDL-PEE-SRB). The suggested framework has two steps: cleaning the information and constructing the model. The information cleansing step utilizes pre-processing methods to manipulate the original information and incorporates lag values as supplementary characteristics. During model-building, the hybrid model undergoes training using the processed data. The HDL model combines fully linked layers, Unidirectional Long Short-Term Memory (Uni-LSTM), and bi-directional LSTMs (Bi-LSTM), which are stacked together. The suggested model is assessed using standard energy usage databases, demonstrating higher accuracy than commonly used HDL models. The proposed model surpasses other HDL models, demonstrating the most favorable performance with a MAPE of 1.94% and an accuracy of 98.06%.