Analysis Of Prediction Of Electrical Power Use Outside Peak Load Of Apartment Building X Using The Long Short Term Memory (LSTM) Method

Abstract

Energy efficiency in residential high-rise buildings has become a critical issue in modern power management, particularly during off-peak periods (LWBP), which contribute significantly to daily electricity consumption. However, most existing studies have primarily focused on peak load forecasting, leaving limited exploration of electricity usage during off-peak hours. This study proposes a daily electricity consumption forecasting model for the off-peak period using the Long Short-Term Memory (LSTM) architecture, designed to capture long-term dependencies in time-series data. The dataset consists of one year of historical daily electricity consumption records from Apartment X. Data preprocessing included Min-Max normalization, time windowing, and partitioning into 80% training and 20% testing sets. Hyperparameter optimization was performed using Optuna, while model performance was evaluated using RMSE, MAE, MSE, and R² metrics. Experimental results demonstrate that the LSTM model effectively captured the temporal patterns of LWBP electricity consumption, achieving RMSE = 0.140, MAE = 0.109, MSE = 0.020, and R² = 0.537. These findings highlight the potential of LSTM as a decision-support tool for building energy management systems, enabling optimization of electricity usage during non-peak hours. Furthermore, this work provides opportunities for future research by integrating hybrid deep learning architectures (e.g., CNN-LSTM or Bi-LSTM) and incorporating external factors such as temperature, weather conditions, and occupant behavior to improve predictive accuracy in real-world applications.