Materials & Energy
Can Existing Meter Data Reveal a Building's Thermal Memory?
2026-09-10
A building's delayed temperature response can suggest a compact thermal model, but several parameter combinations may fit the same meter data. Computational research can test whether the apparent thermal memory is identifiable and transferable. A smooth fitted curve is not enough to justify a control or retrofit claim.
The computational starting point
Use authorized archived indoor and outdoor temperatures, heating input and occupancy proxies where lawfully available. EnergyPlus can provide a reference simulation for controlled checks. Aggregate data to avoid personal occupancy inference. Record whether meter readings represent whole-building input, useful heat or a mixed load with unrelated equipment.
Sources: EnergyPlus input and output reference.
Where intuition enters
A sense that the building stores heat in a hidden layer becomes a two-timescale hypothesis. Compare it with a simpler single-capacitance explanation plus varying gains. Specify which existing temperature transitions should distinguish the models rather than fitting every apparent delay with another hidden state.
A test that can disagree
Fit candidate models on declared periods and evaluate different weather or operating regimes. A bounded numerical fit can explore parameter alternatives, but it cannot make absent excitation appear in the data. Retain multiple near-equivalent solutions and inspect their predictions outside the training schedule.
Check sensitivity to solar gains, internal loads, sensor offsets and meter timing. Use synthetic records to establish whether the fitting procedure can recover known thermal constants under the actual sampling interval. If a parameter changes greatly between equally plausible assumptions, report a range or non-identification rather than a single precise value.
Sources: SciPy nonlinear least squares.
An illustrative decision
Imagine two parameter sets producing almost identical indoor temperatures under a stable heating schedule but divergent cooldown predictions. Existing cooldown records could separate them. Without such periods, the model may predict routine operation adequately while remaining unsuitable for claims about how the building would respond to a changed schedule.
What the research would deliver
The deliverable is a thermal-identifiability map, predictive validation and a boundary on model use. This can guide a building-analytics product's next computational step. It is not a certified energy assessment or an instruction to alter heating controls. A direction preview identifies the most informative archived contrast before broader work is commissioned.
Questions this raises
Does more frequent sampling guarantee identification?
No. The system also needs informative variation and sufficiently documented inputs. More copies of the same regime may add little.
Can this use anonymized building records?
Yes, if aggregation preserves the variables needed for the stated model comparison and the data are authorized for the work.
Sources and their limits
- EnergyPlus input and output reference. Building-system model definitions, not guaranteed equipment performance.
- SciPy nonlinear least squares. Numerical fitting, not physical identification.
Prepared with AI assistance. The linked sources support the specified technical points; they do not validate applied psionics as a whole or guarantee a result for a client.
Read the editorial and evidence standard.
Continue reading
Explore Scientific Oracle consultingfor a scoped review of an existing-data research decision. Start with a non-confidential outline of the question, available evidence and the decision it needs to inform.