Optimization of Residual Hybridization in AIoT-Based Load Forecasting Using LSTM+XGBoost Model

The rapid growth of electricity demand in large institutions and smart campuses calls for accurate short-term load forecasting and real-time monitoring to enable proactive energy management. Conventional metering infrastructure typically provides only instantaneous consumption data without predictive capabilities, limiting its usefulness for decision support. This article proposes an Artificial Intelligence of Things (AIoT) framework that integrates a field-deployed smart metering architecture with an advanced hybrid forecasting model based on Long Short-Term Memory (LSTM) networks and Extreme Gradient Boosting (XGBoost). The LSTM component is designed to capture long-term temporal dependencies in the load time series, while XGBoost is trained on LSTM residuals to model remaining short-term nonlinear variations. This paper shows, using the orthogonality principle of least-squares estimation, that this residual hybridization scheme is mean-square optimal under mild assumptions, thereby providing a formal theoretical justification for the proposed architecture. The LSTM hyperparameters are further optimized using Particle Swarm Optimization (PSO), leading to a substantial reduction in mean squared error compared to Ant Colony Optimization (ACO) and baseline tuned configurations. The proposed framework is evaluated on the widely used GEFCom2014 load forecasting benchmark, where the AIoT-enhanced LSTM–XGBoost model consistently outperforms a standalone LSTM across all error metrics, including mean squared error. Statistical significance tests confirm that the hybrid model’s performance gains over the baseline are highly significant at the α = 0.05 level. In addition, a real-world AIoT deployment at Telkom University demonstrates the feasibility of integrating low-cost digital power meters, non-intrusive current sensors, and Long Term Evolution (LTE)-based connectivity into an end-to-end smart campus architecture. The results highlight the potential of residual hybridization and AIoT-enabled analytics to support more accurate, robust, and scalable load forecasting in institutional energy management systems.