⚡ electrical engineering

Reducing Building Heat Demand Through Intelligent Control: A Comparative Simulation Study

This study demonstrates through comparative simulations that a model predictive control strategy prioritizing thermal comfort tracking can reduce building heating demand more effectively than one minimizing heating power, offering a cost-effective alternative to structural retrofitting.

Original authors: Ueli Schilt, Curtis Meister, Philipp Schuetz

Published 2026-06-17

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Ueli Schilt, Curtis Meister, Philipp Schuetz

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The Smart Thermostat Experiment

Imagine your home heating system is like a car driver. For decades, most drivers (traditional heating controllers) have driven based only on the weather outside. If it's cold, they step on the gas; if it's warm, they take their foot off. They don't really look at how fast the car is actually going or how much fuel they are wasting.

This paper asks a simple question: What if we gave the driver a GPS and a map of the road ahead?

The researchers tested two different "smart drivers" (called Model Predictive Controllers, or MPCs). Both drivers have a map (a computer model of the house) and a weather forecast. They both want to keep the house at a comfortable 20°C. However, they have different goals for how they drive:

Driver A (The "Smooth Cruiser"): This driver's main goal is to avoid slamming on the gas pedal. They want to keep their acceleration (heating power) as low and steady as possible. They are afraid of using "too much power" at any single moment.
Driver B (The "Precision Navigator"): This driver's main goal is to stay exactly on the lane line (the target temperature). They don't care if they have to slam on the gas pedal hard for a few seconds to get back on track; they just want to be perfect at staying at 20°C.

The Setup: A Virtual House

Since they couldn't test this on a real house without risking the tenants' comfort, the researchers built a virtual house inside a computer.

They used a detailed "rulebook" (ISO 52016) to simulate how a real house in Zurich, Switzerland, would react to winter weather.
They created a simpler, faster "mental model" (the RC model) that the smart drivers used to make their decisions.
They ran a 6-day simulation of winter weather to see which driver performed better.

The Results: The Surprise Twist

You might think that the driver trying to save energy (Driver A) would use less fuel. Surprisingly, that wasn't the case.

Here is what happened over the 6 days:

Driver A (Smooth Cruiser):
- Strategy: Because they were penalized for using high power, they never used the full strength of the heater (they stayed under 7.1 kW). They tried to be gentle.
- The Problem: Because they were so gentle, the house temperature drifted away from the target. It got a bit too cold, then they added a little heat, then it got a bit too cold again.
- The Cost: Even though they were "gentle," they ended up using more total energy (603 kWh). It's like driving slowly but inefficiently, taking a winding path that burns more gas in the long run.
Driver B (Precision Navigator):
- Strategy: They didn't mind hitting the gas pedal hard (up to 10 kW) for short bursts to quickly fix the temperature.
- The Result: The house temperature stayed very close to the target (20°C).
- The Cost: Despite using "harder" bursts of power, they actually used less total energy (590 kWh). It's like taking a direct, high-speed route that gets you there faster and with less total fuel consumption.

The "Why": The Squared Penalty

Why did the "gentle" driver use more energy? The researchers found the culprit was the math used to judge Driver A.

The computer told Driver A that using a lot of power at once was "very bad" because the math squared the number.

If you use 1 unit of power, the "badness" score is 1.
If you use 10 units of power, the "badness" score isn't 10; it's 100.

Because the score for high power was so scary, Driver A was terrified of using it. This forced them to use a slow, inefficient strategy that kept the house slightly uncomfortable and wasted energy. Driver B, who wasn't afraid of high power, could use "short, sharp bursts" to fix the temperature quickly, which turned out to be more efficient overall.

The Takeaway

The paper concludes two main things:

Smart control works: You can keep a house very comfortable without major structural renovations (like adding insulation) just by using smart software.
The goal matters: How you tell the computer what to optimize changes the result. If you tell the computer to "avoid high power at all costs," it might actually waste more energy. If you tell it to "keep the temperature perfect," it might find a more efficient way to do it.

In short: Being too afraid of using a lot of power right now can actually cost you more energy in the long run.

Technical Summary: Reducing Building Heat Demand Through Intelligent Control

Problem Statement
Space heating constitutes the dominant energy consumption in buildings. While structural retrofitting effectively reduces demand, it is often costly and time-intensive. Existing literature suggests that replacing conventional weather-compensation controllers (heating curve controllers) with intelligent Model Predictive Control (MPC) strategies can significantly reduce heating energy. However, most studies compare MPC against conventional control methods. There is a gap in understanding how the specific formulation of the MPC objective function influences the trade-off between thermal comfort (temperature tracking) and heat demand. Specifically, it is unclear whether prioritizing strict temperature tracking versus minimizing heating power yields different energy outcomes.

Methodology
The study employs a comparative simulation approach using a virtual building model to generate high-resolution synthetic data.

Virtual Plant: A detailed building thermal model based on ISO 52016-1 was implemented in Python to simulate a two-story, 100 m² residential building in Zurich, Switzerland, over a three-month period (January–March 2020). This "ISO model" generated synthetic measurement data using weather inputs and a simple linear heating controller.
Internal Controller Model: A simplified single-node Resistance-Capacitance (RC) model (1-R-1-C) was parametrized using the synthetic data from the ISO model. This RC model serves as the internal plant model for the MPC controller, implemented in MATLAB.
MPC Simulation: Two distinct MPC strategies were simulated over a 144-hour (6-day) period using the MATLAB MPC Toolbox. Both strategies utilized the same prediction horizon (12 hours), control horizon (12 hours), and physical constraints (heating power limits, temperature bounds). They differed solely in their quadratic objective function weights:
- Simulation A (Power-Minimizing): Prioritized minimizing heating power ( $w_{\dot{Q}_h}=1, w_{T_i}=0$ ).
- Simulation B (Comfort-Tracking): Prioritized tracking the reference room temperature ( $w_{\dot{Q}_h}=0, w_{T_i}=1$ ).
- Both maintained a reference temperature of 20.0°C (18.0°C at night) and respected comfort bounds ( $T_{ref} \pm 0.5^\circ\text{C}$ to $+5.0^\circ\text{C}$ ).

Key Results
The simulation results over six days revealed distinct performance characteristics between the two strategies:

Thermal Comfort: Simulation B (comfort-tracking) achieved superior temperature regulation with a Root-Mean-Squared Error (RMSE) of 0.30°C, compared to 0.82°C for Simulation A.
Heating Power Dynamics: Simulation A avoided extreme heating power values, never exceeding 7.1 kW. In contrast, Simulation B frequently utilized the maximum allowed heating power of 10 kW to correct temperature deviations quickly.
Energy Consumption: Counter-intuitively, Simulation A (minimizing power) resulted in a higher cumulative heating energy demand (603 kWh) compared to Simulation B (590 kWh). Simulation B achieved a 2% reduction in total energy use despite utilizing higher peak power.
Constraints: Both strategies successfully adhered to all defined heating power and temperature constraints.

Significance and Claims
The paper claims that the choice of objective function formulation is a critical determinant of system performance in MPC-based building control.

Novelty: The study highlights that while MPC is typically formulated to minimize cost or energy, applying it solely for setpoint tracking (Simulation B) is an uncommon approach that yielded unexpected results.
Mechanism of Energy Savings: The authors attribute the lower energy consumption of the comfort-tracking controller to the quadratic nature of the power-minimization objective. By squaring the heating power term, the cost function disproportionately penalizes high power rates, forcing the controller to avoid extreme inputs. This avoidance leads to a more gradual, less efficient heating profile that ultimately consumes more total energy. Conversely, the comfort-tracking controller utilizes high-power bursts to maintain tight temperature control, which proves more energy-efficient in this specific simulation context.
Implications: The results suggest that MPC can achieve high thermal comfort levels without compromising energy demand, and potentially even reducing it, without structural changes to the building envelope. However, the authors modestly note that these findings are specific to the simulated virtual building and ambient conditions, and that the influence of objective function formulation (e.g., using non-quadratic objectives or direct energy minimization) warrants further investigation.