The intelligent system known as the Smart Grid is now recognized as an essential component for the power grid's reliable operation. Because of rising global energy demand and growing environmental concerns, energy conservation and sustainable energy sources are becoming increasingly crucial. Energy supply and management are especially difficult in energy-intensive sectors such as large-scale construction. These constructions may have significant energy demands at times, but they may also have difficulties with energy waste. Predicting power consumption in smart grids is critical for addressing this issue. In this paper, the ensemble Deep Learning (DL) model is developed for accurate forecasting of power consumption. The ensemble network known as HFCM (Hierarchical Feature Coupled Module)-LSTM (Long-Short-Term Memory) is used to estimate power consumption more accurately. This network is designed to automatically extract the important features from the input to predict the output. To assess the model's efficiency, traditional DL models such as Convolutional Neural Network (CNN) and LSTM are employed. The acquired and preprocessed smart grid data from Kaggle is used. The recommended ensemble model achieves the highest R2 score of 0.986 in power prediction, while the CNN and LSTM reach 0.934 and 0.969, respectively. In addition to R2, various error measures are used for validation. Based on the results of the experiments, the proposed model beats other models with much lower error rates. © 2024 Elsevier B.V., All rights reserved.