Earticle

초록

한국어

The rapid expansion of generative AI and cloud computing is transforming industrial structures and driving unprecedented growth in data centers. As a result, electricity demand is becoming increasingly shaped by digital technologies rather than traditional climate or economic factors alone. However, existing forecasting models largely overlook these sociotechnical drivers, risking substantial underestimation of future demand. This study integrates technology diffusion indicators with climate variables to examine whether digital-technology trends meaningfully contribute to national electricity demand. Using a Double Machine Learning framework as a feature-validation step, we confirm that GPT search volume and cloud market size serve as statistically robust and predictive indicators of electricity consumption. In addition, Fourier Transform–based features are employed to capture periodic variability, significantly improving forecasting performance beyond climate- and economy-centric baselines. Scenario simulations under the SSP585 pathway forecast a steady rise in demand through 2045, with seasonal peaks amplified by AI and cloud adoption. The findings highlight the structural role of sociotechnical factors in shaping electricity demand and offer practical implications for electricity pricing, infrastructure planning, and risk management.

목차

ABSTRACT
Ⅰ. Introduction
Ⅱ. Conceptual Background
2.1. Electricity Demand Forecasting in the Age of AI and Cloud
2.2. Literature Review
Ⅲ. Research Methodology
3.1. Research Procedure
3.2. Double Machine Learning
3.3. Forecasting Model
3.4. Data Collection
Ⅳ. Result
4.1. Results of Empirical Validation of Predictive Relevance
4.2. Results of Fourier-Based Periodicity analysis
4.3. Lag Feature Engineering and Normalization
4.4. Comparison of Forecasting Performance
4.5. Feature Importance
4.6. Scenario-Based Long-Term Forecasting Simulation
Ⅴ. Discussion and Implications
5.1. Findings and Proposed Strategies
5.2. Implications for Research and Practice
5.3. Limitations and Future Research Directions
Acknowledgements

저자

• Jinho Kang [ Master's Candidate, Graduate School of Information, Yonsei University, Korea ]
• Min-ho Song [ Master's Candidate, School of Industrial & System, Kyonggi University, Korea ]
• So-Hyun Lee [ Assistant Professor, Department of Industrial and Management Engineering, Kyonggi University, Korea ] Corresponding Author
• Hee-Woong Kim [ Professor, Graduate School of Information, Yonsei University, Korea ]

참고문헌