<![CDATA[The implementation of the Work from Anywhere (WFA) system has shifted electricity consumption patterns from centralized office environments to distributed locations such as homes, personal devices, co-working spaces, and internet-based digital services. In Indonesia in 2026, this phenomenon has become increasingly visible due to the growth of online activities and household electronic device usage, which significantly affect national electricity demand. In practice, several institutions and universities have started facing challenges in measuring energy efficiency because electricity consumption is no longer concentrated in a single location. This study proposes an Artificial Intelligence-based deep learning approach to predict energy efficiency levels in WFA environments using multivariate synthetic data. The dataset was constructed from several variables, including daily working duration, number of active devices, room temperature, occupancy level, air conditioning usage, internet connection quality, electricity tariffs, and electricity emission intensity. Four models were evaluated, namely Random Forest as the baseline model, Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and a lightweight Transformer model. Simulation results indicate that the lightweight Transformer achieved the best performance with an MAE of 0.041, RMSE of 0.057, MAPE of 5.31%, and a coefficient of determination (R²) of 0.928. The evaluation values were obtained from simulation results of the models using synthetic WFA system data. Overall, the findings demonstrate that deep learning approaches have strong potential to support data-driven monitoring systems and decision-making processes for improving energy efficiency in modern digital working environments]]>
Copyrights © 2026
