AI for heat pumps: combining solar gains, thermal inertia and tariffs

A weather compensation curve, domestic hot water priority and an operating schedule do not require a separate AI platform. Some manufacturers also offer weather forecasts, photovoltaic integration and adaptation based on building data. NIBE, for example, describes these functions. (NIBE — air source heat pumps: weather forecasts and solar generation integration)

So we do not start by claiming that the built-in controls cannot do anything. We start with a specific question: which decision could be made better for this property?

Solar forecasting: making better use of useful heat

The sun may warm rooms after the heating has already stored heat in the floor. A predictive control strategy reduces heating in advance so that solar gains can meet part of the demand.

But AI cannot take credit for all solar energy used. Conventional controls will also reduce heating once the rooms have warmed up. What needs to be calculated is the additional benefit of forecasting compared with this baseline operation.

For illustration, assume that 4 kWh of solar heat is usefully used without forecasting, compared with 7 kWh with forecasting. If the 3 kWh difference genuinely replaces heat from the heat pump at a COP of 3, the calculated reduction in electricity consumption is 1 kWh.

This is the result of a defined scenario, not a seasonal guarantee. The comparison must maintain the agreed comfort levels and account for subsequent reheating. It must also compare the building’s final thermal state or account for the difference in stored energy.

Reducing overheating by 2 °C does not, on its own, establish how many kilowatt-hours have been saved: that requires an energy balance, not an arbitrary heat capacity per square metre.

Night-time tariffs: a lower bill does not always mean less energy

Manufacturers already use price forecasts to shift the timing of heat production. NIBE describes this approach in Smart Price Adaption, subject to a suitable tariff contract. (NIBE S735 — Smart Price Adaption)

Let us calculate our own hypothetical scenario. The building needs 20 kWh of useful heat. The daytime COP is 3,8 and the night-time COP is 3,2. For heat stored in advance, assume that 90% of the heat produced is usefully used. The prices are illustrative, not current tariffs: 0,24 €/kWh during the day and 0,12 at night.

Heat produced directly during the day:

E_d=20/3,8=5,26 kWh; C_d≈1,26 €

Night-time heat production with the assumed storage losses:

E_n=20/(3,2×0,9)=6,94 kWh; C_n≈0,83 €

For this quantity of heat, costs fall by approximately 34%, although electricity consumption rises by approximately 32%. There is no contradiction: cost optimisation and energy optimisation have different criteria.

Solar gains, tariffs and photovoltaics may call for different actions

If strong solar gains are expected tomorrow, additional night-time heat storage may be unnecessary. If there is surplus on-site electricity generation during the day, shifting production to the night is not necessarily the most cost-effective option.

Our task is to compare the scenarios together, not add up unrelated marketing percentages. This requires data on the building’s response, available setpoints, temperature limits, energy sources and user priorities.

What the client receives

As part of AI Engineering, we would first build a digital context for the property and test recommendations without automatic intervention. Next would come agreed changes to setpoints and schedules, with a decision log. Factory safety functions and basic controls remain independent.

Results should be reported separately: energy saved, cost reductions, comfort and peak load. In this approach, AII Systems is an implementation tool, not a substitute for engineering calculations.

The value of AI is not that it changes the temperature more often. It is that every change has a clear reason, acceptable limits and a verifiable outcome.