Weather forecasting for MPC can influence how effectively a building control system balances energy use with indoor comfort. Model predictive control (MPC) plans control actions using predictions of future conditions, but conventional weather forecasts may not be well suited to short horizons. A disturbance model based on local weather observations offers another approach: it estimates how conditions such as solar radiation and ambient temperature are likely to change and supplies those estimates to the controller.
A simulation study of building climate control found that this approach could improve comfort and reduce electricity use compared with MPC using persistent disturbance estimates. The results are specific to the simulated buildings and conditions, but they highlight an important engineering principle: forecast quality matters, particularly when disturbances change faster than the heating system can respond.
Why weather disturbances matter to building MPC
In building climate control, solar radiation and outdoor air temperature affect indoor temperature in different ways. Sunlight entering through windows can add heat quickly, while outdoor temperature influences heat transfer through the building envelope. These disturbances can change on timescales that complicate temperature regulation.
MPC uses a model of the building and a forecast of future disturbances to select control inputs over a prediction horizon. It can account for operating constraints and an objective such as minimizing electricity cost. At each control step, it updates its state estimate and solves the optimization problem again. If the disturbance forecast assumes that current conditions will remain unchanged, however, the controller may not anticipate a coming shift in solar gain or outdoor temperature.
Offset-free MPC addresses certain modeling errors by estimating an unknown, persistent disturbance. That can help compensate for a steady mismatch, but a constant estimate does not describe a rapidly changing weather signal. The study therefore investigated whether more detailed short-term forecasts could improve conventional, certainty-equivalent MPC without requiring a stochastic MPC formulation.
Weather forecasting for MPC with disturbance models
The proposed method represents both the building and weather disturbances using stochastic differential equations (SDEs). SDEs describe how system variables evolve over time while also representing process uncertainty. In the building model, relevant states include room-air temperature and, depending on the heating configuration, the temperatures of the floor and heating-system water. The weather model focuses on ambient air temperature and solar radiation.
Rather than treating weather as a single, unrelated input, the forecasting approach builds estimates from connected weather variables. Its components include cloud cover, global solar radiation, net radiation and ambient air temperature. For example, cloud cover helps explain changes in solar radiation, while net radiation contributes to the modeled dynamics of outdoor air temperature. The combined model uses these relationships to generate a sequence of short-term disturbance estimates for the MPC prediction horizon.
Why the Lamperti transformation is used
Some variables, including cloud cover, have a bounded range. A stochastic model with state-dependent uncertainty can be difficult to estimate and simulate reliably if numerical calculations produce values outside that range. The study applies a Lamperti transformation to map the bounded cloud-cover process into a domain where the transformed model has constant diffusion. Forecasting and estimation can then be carried out in that domain before results are mapped back to the original variable.
This is a modeling and computation choice, not a change to the physical meaning of cloud cover. It helps make the SDE-based method practical within the forecasting and filtering framework.
How the forecast connects to the MPC controller
The control architecture separates weather forecasting from building-state estimation. Local measurements inform estimates of the present weather and indoor conditions. The disturbance model then supplies forecasts of ambient temperature and solar radiation. MPC uses those forecasts, the estimated building state and the system model to calculate a sequence of heating or cooling inputs while respecting input and temperature constraints.
In the evaluated setup, the controller optimizes electricity use while penalizing violations of the desired indoor-temperature range. It recalculates the control action when new measurements arrive. This structure is relevant to energy management systems because it allows a disturbance model and an MPC controller to operate as distinct components, provided their data and model interfaces are coordinated.
Heating-system dynamics affect the value of better forecasts
The simulations compared several configurations, including electric heaters, a heat pump, and combinations of heating and cooling equipment. The modeled heat pump was more energy efficient than the faster electric heaters, but its response was slower. That difference matters: a slow actuator may need more advance notice of changing conditions, while a faster heater can respond more readily to a sudden disturbance.
In the study, advanced forecasts made a particularly visible difference for the heat-pump-only configuration. Combining the heat pump with faster electric heating also allowed the controller to use the efficient heat pump for slower heating needs and respond more quickly when required. These findings illustrate why forecast design should be considered alongside actuator dynamics, thermal storage and operating constraints—not as a standalone software feature.
What the simulation results show—and what they do not
Compared with persistent disturbance forecasts used in offset-free control, the advanced short-term forecasts reduced simulated comfort-constraint violations across the tested heating strategies. The paper reports electricity savings in the range of 5%–10% and reductions in constraint violations of up to more than 90%. Some results approached those obtained using perfect forecasts, which serve as a theoretical comparison rather than an operationally available input.
The evidence comes from simulations of building models using historical Danish weather data, not from a reported deployment in a live facility. The statistical results for constraint violations were stronger than those for electricity cost: the authors note that the cost reductions did not reach usual significance levels in their tests. The results should therefore be treated as an indication of potential under the studied conditions, not a guaranteed outcome for other buildings, climates or control configurations.
The authors also note that advanced local models are intended for short-term forecasting. Meteorological forecasts may be more accurate at longer horizons, so combining local short-term predictions with external forecasts is a promising direction, but the best transition point requires further investigation.
Practical implications for automation teams
For engineers and system integrators evaluating MPC for building or energy applications, the study points to several practical considerations:
- Match forecasts to the control horizon. Short-term local forecasts and longer-range meteorological forecasts may serve different roles.
- Account for actuator response. Slow equipment may benefit more from forecasts that anticipate disturbances, while faster devices can handle some changes after they occur.
- Model relevant disturbances. Solar radiation and ambient temperature can affect indoor conditions through different mechanisms.
- Validate performance in context. Test forecast and control strategies against the building model, available measurements, comfort limits and energy objectives relevant to the application.
Embedding weather disturbance models in MPC offers a structured way to bring short-term local forecasts into building climate control. The reported simulations suggest potential improvements in comfort and electricity use compared with persistent disturbance estimates, while also showing that equipment dynamics shape the benefit. For plant and building automation teams, the practical takeaway is to evaluate forecasting, control, sensing and actuator behavior as parts of one integrated system—and to validate claimed gains under the conditions the system will actually face.
Access Official Technical PDF