Scholarlink

<p>The constantly-increasing intricacy of household energy consumption in smart cities increases the demand for accurate forecasting models that will enable efficient energy management and sustainability. Most traditional statistical methods fail to model nonlinear temporal and electrical dependencies in real-world IoT data. An integrated IoT-enabled smart energy management framework is developed that couples advanced machine learning and deep learning models with a metaheuristic optimization algorithm. Real-world IoT data from the UCI household power consumption dataset were utilized, where ensemble-based models (XGBoost, LightGBM, HGB) and neural network architectures (MLP and LSTM) were optimized using the Golden Jackal Optimizer (GJO). Among all models, the optimized GJO-MLP model obtained the highest predictive performance with an RMSE of 0.222 and R<sup>2</sup> of 0.946, while the GJO-LightGBM model came in second. SHAP-based feature interpretation identified short-term lags of global active power and voltage as the dominant predictors. This framework enhances transparency, interpretability, and predictive accuracy in real-time demand forecasting. This work has shown that AI-driven IoT energy management systems can be used to make significant improvement in demand-side management, energy efficiency, and operational stability in future smart grids.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

IoT-Enabled Smart Urban Energy Management Using Optimized Ensemble-Based and Neural Network Architectures for Household Consumption Forecasting

  • 期刊论文

  • 发表日期：

    2026年4月30日

查看全文
  • Tiequan Chen

摘要

The constantly-increasing intricacy of household energy consumption in smart cities increases the demand for accurate forecasting models that will enable efficient energy management and sustainability. Most traditional statistical methods fail to model nonlinear temporal and electrical dependencies in real-world IoT data. An integrated IoT-enabled smart energy management framework is developed that couples advanced machine learning and deep learning models with a metaheuristic optimization algorithm. Real-world IoT data from the UCI household power consumption dataset were utilized, where ensemble-based models (XGBoost, LightGBM, HGB) and neural network architectures (MLP and LSTM) were optimized using the Golden Jackal Optimizer (GJO). Among all models, the optimized GJO-MLP model obtained the highest predictive performance with an RMSE of 0.222 and R2 of 0.946, while the GJO-LightGBM model came in second. SHAP-based feature interpretation identified short-term lags of global active power and voltage as the dominant predictors. This framework enhances transparency, interpretability, and predictive accuracy in real-time demand forecasting. This work has shown that AI-driven IoT energy management systems can be used to make significant improvement in demand-side management, energy efficiency, and operational stability in future smart grids.