Share this
Why Real-World Models Must Factor in Humidity, Thermal Lag, and Cooling Efficiency in Power Forecasting
Power load forecasting can be as simple as correlating demand with temperature or as sophisticated as AI or machine-learning ensembles. Each approach has pros and cons, but they all risk missing the deeper drivers of power demand volatility.
At Yes Energy, we rely on a time-varying coefficient regression model. It captures not only the obvious temperature signal but also the harder-to-see effects – humidity, overnight cooling, building thermal mass, and more – that pure “temperature-onlyˮ or “last-yearˮ models often overlook.
Let’s see it in action.
June 2025 delivered record-breaking heat across Japan. The time frame provides a perfect case study to show where different models shine – or stumble.
Why Temperature-Only Models Come Up Short
A model that maps temperature (or any single weather input) to power demand will struggle to explain complex weather swings like those we just experienced.
When operators canʼt “seeˮ why the predictions changed, confidence drops and the quality of decision-making suffers.
Case Study: Tokyo, Summer 2025, Mid-June (Observed) vs. Early July (Predicted)
The table below compares three-day periods built from observed June data and projected July data for Tokyo. “Average Daytime Tempˮ is the mean from 10:00 to 16:00.
Source: Yes Energy
July demand is predicted to be ≥ 1.5 GW higher, even though June was hotter. A simple temperature-to-demand curve canʼt explain the gap.
Whatʼs Driving the Power Forecasts’ Difference?
- Humidity packs a bigger punch than you think. For peak hours in July, humidity is projected to run 6 to 9 percentage points higher than in June.
Higher humidity:
- Raises perceived temperature
- Extends AC run times and duty cycles
- Pushes total load upward.
- Thermal lag and the cooling feedback loop contribute. Consecutive hot and humid days in early July mean nights donʼt cool off, and buildings start the day already warm. AC units ramp early and hard, and efficiency falls as humidity rises.
In June, cooler nights allowed structures to shed heat, keeping the next-day load lower.
Visual Evidence: Demand vs. Weather Overlays
Temperature and Demand
Source: Yes Energy Demand Forecasts
Blue represents June (actual). We see higher temperatures, but lower demand. Green represents July (projected). We predict slightly lower temperatures but higher demand.
Humidity and Demand 
Source: Yes Energy Demand Forecasts
Orange represents June with 55 to 61% humidity. Red represents July with 62 to 65% humidity. Higher humidity correlates with increased demand.
About the Simulation Forecast
Forecast: Baseline demand used the official humidity outlook (see charts above).
Simulation: Demand recalculated with humidity 6% lower on July 8.
Difference: The direct impact of humidity change (lag effects excluded).
Even a small relative humidity tweak can shift peak load by a full gigawatt – proof that multi-factor models beat “temperature-onlyˮ models every time.
We also ran a what-if simulation: What if July 8 humidity dropped by the same 6%?
The result: nearly 1 GW lower peak demand, which confirms humidityʼs outsized role. A 1.52 GW swing can move market prices dramatically.
Source: Yes Energy
What Makes the Yes Energy Model Different?
For Japan, North America, and Europe, our Yes Energy model:
- Integrates temperature, humidity, overnight cooling loss, and more
- Accounts for short-term lag effects between weather shifts and load
- Provides transparent visuals so you can see why the projection moves.
During the summer peak, you need power load forecasts that mirror reality – not just the thermometer. Our team also answers your questions, and unlike with AI power forecasting models, we can explain the “why” behind the forecast, giving you increased confidence.
Talk to our team to see our power load forecasting in action.
About the author: Tetsuya Totsuka is an APAC power-market expert covering Japan, Australia, and New Zealand, specialising in electricity-demand forecasting and statistical modelling. With more than a decade of technical-sales and consulting experience, he turns wholesale-market complexity into clear, actionable strategy for Yes Energy’s clients through data-driven insights.
Share this
- Industry News & Trends (110)
- Power Traders (83)
- Asset Managers (44)
- Data, Digital Transformation & Data Journey (35)
- Asset Developers (31)
- ERCOT (31)
- Market Events (29)
- PowerSignals (28)
- Infrastructure Insights Dataset (27)
- Utilities (27)
- Market Driver Alerts - Live Power (25)
- Yes Energy Demand Forecasts (24)
- DataSignals (22)
- Live Power (22)
- ISO Changes & Expansion (20)
- Renewable Energy (19)
- Risk Management (18)
- Data Scientists (17)
- CAISO (15)
- Energy Storage / Battery Technology (15)
- PJM (14)
- EnCompass (11)
- QuickSignals (11)
- SPP (11)
- MISO (9)
- Position Management (9)
- Power Markets 101 (9)
- Submission Services (8)
- Financial Transmission Rights (7)
- Data Centers (6)
- Snowflake (6)
- Powered by Yes Energy (5)
- Asset Developers/Managers (4)
- Demand Forecasts (4)
- FTR Positions Dataset (4)
- Geo Data (4)
- ISO-NE (4)
- Solutions Developers (4)
- Battery Operators (3)
- Commercial Vendors (3)
- IESO (3)
- Independent Power Producers (3)
- NYISO (3)
- data quality (3)
- AI and Machine Learning (2)
- Canada (2)
- Europe (2)
- GridSite (2)
- PeopleOps (2)
- Crypto Mining (1)
- FERC (1)
- Ireland (1)
- Japanese Power Markets (1)
- Natural Gas (1)
- PowerCore (1)
- Western Markets (1)
- hydro storage (1)
- nuclear power (1)
- June 2025 (4)
- May 2025 (5)
- April 2025 (10)
- March 2025 (6)
- February 2025 (11)
- January 2025 (7)
- December 2024 (4)
- November 2024 (7)
- October 2024 (6)
- September 2024 (5)
- August 2024 (10)
- July 2024 (9)
- June 2024 (4)
- May 2024 (7)
- April 2024 (6)
- March 2024 (4)
- February 2024 (8)
- January 2024 (5)
- December 2023 (4)
- November 2023 (6)
- October 2023 (8)
- September 2023 (1)
- August 2023 (3)
- July 2023 (3)
- May 2023 (4)
- April 2023 (2)
- March 2023 (1)
- February 2023 (2)
- January 2023 (3)
- December 2022 (2)
- November 2022 (1)
- October 2022 (3)
- September 2022 (5)
- August 2022 (4)
- July 2022 (3)
- June 2022 (2)
- May 2022 (1)
- April 2022 (3)
- March 2022 (3)
- February 2022 (6)
- January 2022 (3)
- December 2021 (1)
- November 2021 (2)
- October 2021 (4)
- September 2021 (1)
- August 2021 (1)
- July 2021 (1)
- June 2021 (2)
- May 2021 (3)
- April 2021 (3)
- March 2021 (3)
- February 2021 (3)
- December 2020 (3)
- November 2020 (4)
- October 2020 (2)
- September 2020 (3)
- August 2020 (2)
- July 2020 (2)
- June 2020 (1)
- May 2020 (8)
- November 2019 (1)
- August 2019 (2)
- June 2019 (1)
- May 2019 (2)
- January 2019 (1)