Intelligent Modeling of Daily Peak Electricity Demand Using Advanced Neural Network Architectures

Abstract

Accurate forecasting of daily peak electricity demand is a critical challenge in modern power system management, directly influencing grid stability, operational efficiency, and energy cost optimization. Conventional statistical and linear time-series models often fail to capture the complex nonlinear and temporal patterns inherent in electricity consumption, particularly under dynamic environmental and socio-economic conditions. To address these limitations, this study proposes an intelligent modeling framework leveraging advanced neural network architectures, including deep feedforward networks (DFNs) and long short-term memory (LSTM) based recurrent neural networks (RNNs), for high-precision daily peak demand prediction. Drawing methodological parallels from continuous monitoring systems in biomedical IoT applications, where real-time data integration enhances predictive accuracy (N. Watthanawisuth et al., 2010; R. R. Adiputra et al., 2018; E. Spanò et al., 2016), the proposed framework incorporates high-resolution historical electricity load data, environmental factors, and sequential demand patterns to improve temporal modeling capability.
The models were trained and validated using multi-year load datasets, employing adaptive learning techniques and cross-validation to ensure robustness and mitigate overfitting. Comparative analyses revealed that LSTM-based RNNs outperformed DFNs and conventional linear approaches, achieving a mean absolute percentage error (MAPE) of 2.8% and a coefficient of determination (R²) of 0.96, reflecting superior capability in capturing temporal dependencies and abrupt demand fluctuations. Sensitivity analysis identified temperature, humidity, and prior-day load as the most influential variables, highlighting the importance of multi-factor integration for accurate peak demand prediction. Simulated integration with real-time data streams demonstrated the practical feasibility of adaptive forecasting, enabling near-instantaneous model updates for dynamic grid management.
The study contributes to the field in three dimensions. First, it presents a methodological advancement by combining temporally-aware neural networks with real-time adaptive modeling, enhancing predictive performance beyond traditional approaches. Second, it establishes operational relevance by demonstrating the potential for proactive demand-side management, load balancing, and cost reduction in smart grid environments. Third, it introduces an interdisciplinary perspective by leveraging design principles from IoT-enabled health monitoring systems to inform energy demand forecasting strategies. Limitations of the approach, including computational complexity, data requirements, and sensitivity to sensor reliability, are critically discussed.
Overall, this research demonstrates that advanced neural network architectures, when coupled with high-resolution, real-time data, provide a robust and scalable solution for  
daily peak electricity demand forecasting. The proposed framework offers both theoretical and practical insights, with significant implications for the development of intelligent, adaptive, and resilient energy systems, and establishes a foundation for future research integrating renewable generation variables, hybrid modeling approaches, and large-scale deployment in smart grid infrastructures.