Economic and environmental analysis of forecasting errors in air conditioning energy demand under fixed and time-of-use tariffs: A case study

Accurate forecasting of Air Conditioning (AC) energy demand is essential for mitigating economic and environmental challenges in energy planning, especially under extreme weather conditions. Despite advancements in prediction techniques, limited attention has been given to the real-world impacts of forecasting errors under different electricity tariff structures. This study quantifies these impacts by modeling a typical office space using the Hourly Analysis Program (HAP) and evaluating 8 machine learning models. Gradient Boosting Regressor (GBR) achieved the highest accuracy (R2 = 95.004%, RMSE = 0.0883 kWh) and was selected for subsequent economic and environmental analyses. Fixed and Time-of-Use (TOU) tariffs were assessed across three cases: (1) no penalties, (2) penalties for underestimation, and (3) penalties for both under- and overestimation. Sensitivity analysis showed that increasing underestimation penalties from 0.03 to 0.15 USD/kWh led to cost increases of 1.9%-10.6% under fixed tariffs and 2.0%-11.2% under TOU tariffs. Similarly, increasing overestimation penalties in the same range resulted in cost increases of 4.3%-8.6% for fixed tariffs and 4.5%-9.1% for TOU. While percentage increases were slightly higher under TOU, their normalized costs were consistently lower, resulting in lower absolute costs overall. TOU tariffs thus continued to favor commercial AC use, offering greater resilience to forecast inaccuracies. Seasonal analysis revealed higher winter penalties, while overestimations caused a 6.00% increase in CO2 emissions. The single-office case study offers a controlled setting to isolate forecasting impacts, with normalized metrics ensuring scalability. We find that it is important to integrate forecasting accuracy into tariff design and energy policy to avoid economic losses and environmental harm.