The growing adoption of smart homes into the contemporary energy systems has aggravated the necessity of precise energy consumption forecasting to complement the demand-side management, peak load decrease, and integration of renewable energy. Conventional forecasting techniques such as ARIMA and support vector regression do not tend to capture non-linear and time-dependent relationships of household energy consumption data. To overcome these issues, this paper suggests a Long Short-Term Memory (LSTM) network based predictive model. They used a publicly available smart home dataset, and the preprocessing steps used were data normalization, feature engineering, and extraction of temporal lags. The LSTM architecture was trained and tested as compared to baseline models and the performance measured in terms of RMSE, MAE, MAPE and R 2. It was found that the LSTM model had better accuracy with the forecasting error decreasing by about 2025% relative to the classical process. The results demonstrate a promising future of deep learning in intelligent energy systems and offer a powerful means of efficient energy scheduling and a sustainable grid operation. The suggested framework possesses scalability and flexibility to use in real-world application, which is the foundation of integration into demand response plans and IoT-based smart environments.