Answers are grounded in this paper and cite their sources.
Buildings account for 36–40% of global energy consumption and CO₂ emissions, with electricity representing nearly 55% of global electricity use. Accurate long-term forecasting is critical for grid stability and net-zero targets, yet most predictive models require extensive multi-year datasets or complex machine learning architectures. To our knowledge, no prior study has employed Fourier series methods integrating climate variables and weekday residual corrections to predict full annual electricity consumption using only a single year of training data, representing a methodological gap for long-term, one-year-ahead prediction in data-scarce environments. This study develops a simple modeling framework using Fourier series enhanced by multivariate regression incorporating climate variables (air temperature, humidity, sunshine hours) and seasonal residual correction. The developed models were tested on 322 apartments in a residential building in Japan, using hierarchical clustering to identify distinct consumption clusters (low, medium and high). The model achieved high accuracy across all clusters in this case study (RMSE = 0.472–1.20 kWh, MAPE = 7.1–8.9%) when forecasting one year ahead. Autoregressive residual correction further improved day-ahead forecasting accuracy (RMSE = 0.49–0.83 kWh, MAPE = 4.8–5.5%). Residual analysis revealed systematic weekday–weekend patterns and strong correlations with temperature. High-consumption clusters exhibited two to three times greater temperature sensitivity than low-consumption groups. The proposed Fourier approach provides an interpretable, data-efficient alternative to complex machine learning models, enabling accurate one-year-ahead electricity forecasting using only a single year of historical data. It offers a practical tool for building energy managers and policymakers operating in data-constrained contexts.
No takes yet. Share an insight, caveat, or question.
Appuhamilage et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: