MPC for HVAC Energy Efficiency — PatSnap Eureka
How MPC Improves HVAC Energy Efficiency in Smart Buildings
Model Predictive Control transforms HVAC from reactive setpoint chasing to proactive, constraint-aware optimization — delivering up to 19.4% energy savings while maintaining occupant comfort, based on analysis of 60+ patent filings from 2012–2026.
From Reactive Thermostats to Proactive Optimization
Model Predictive Control (MPC) represents a fundamental shift in how building energy systems are managed. Rather than reacting to temperature deviations after they occur, MPC uses a thermal model to predict future zone temperatures over a defined horizon, runs an optimizer to minimize a cost function subject to comfort and equipment constraints, and applies only the first control action — before repeating the entire cycle at the next time step.
This receding-horizon strategy, documented across more than 60 patent filings spanning the US, EU, WO, JP, KR, CN, AU, and DE jurisdictions, enables HVAC systems to preemptively counteract anticipated thermal loads rather than reacting to them after the fact. According to the U.S. Department of Energy, buildings account for approximately 40% of total U.S. energy consumption — making intelligent HVAC control a critical lever for sustainability.
The consistent finding across the patent corpus is that MPC outperforms traditional rule-based or fixed setpoint control by enabling proactive, constraint-aware energy minimization over a rolling time horizon. Research from Xi'an University of Architecture and Technology quantifies this advantage: MPC achieves energy savings of up to 19.4% versus conventional on/off control while keeping PMV (Predicted Mean Vote) comfort indices within acceptable ranges, with approximately 6% savings when comfort is the primary objective.
For R&D teams and IP professionals tracking this space, patent landscape analytics reveal a field dominated by vertically integrated industrial players — Johnson Controls, Bosch, Honeywell, Siemens — supplemented by academic institutions and software-first startups like BrainBox AI.
Thermal Modeling Frameworks: The Foundation of MPC Performance
The effectiveness of any MPC scheme for HVAC hinges on the fidelity and adaptability of the underlying building thermal model. Four distinct approaches dominate the patent corpus.
Physics-Based Cascaded Models
Johnson Controls Technology Company's 2021 patent describes a cascaded architecture that separates a disturbance model — predicting heat disturbances from solar radiation and occupancy — from a physics model predicting zone temperature as a function of that disturbance and HVAC output. A combined training procedure simultaneously estimates parameters of both components, enabling MPC to preemptively counteract anticipated thermal loads.
Proactive load anticipationState-Space Formulations with MHE
Palo Alto Research Center's 2023 patent uses geometric metadata such as floor plans and room dimensions to initialize a base thermal model, then employs Moving Horizon Estimation (MHE) to iteratively fit model parameters to observed time-series data. The resulting updated model is used for MPC-based receding horizon control, explicitly balancing energy efficiency and occupant comfort.
Geometry-initialized MHEAdaptive Model Correction
ABB AG's 2019 patent addresses model drift: the MPC system continuously measures zone temperature and its rate of change, compares these against model predictions, identifies probable causes of error with associated probabilities, evaluates their impacts, and updates the building thermal model accordingly. This closed-loop model refinement maintains prediction accuracy over the building's operational lifetime.
Probabilistic error diagnosisScalable Embedded MPC
Palo Alto Research Center's 2024 patent linearizes a physics model around an operating point and uses the linearized model as an equality constraint for the MPC optimizer. The constraint matrix is factored into U and V matrices such that the relevant matrix products are diagonal, enabling iterative solution via efficient linear algebra rather than dense matrix inversion — substantially reducing computation time for long prediction horizons.
Edge-deployable computationMPC Performance and Patent Landscape at a Glance
Key metrics extracted from 60+ patent documents spanning 2012–2026, illustrating energy savings potential and the distribution of innovation activity across assignees.
MPC Energy Savings vs Conventional Control
Up to 19.4% savings in energy-primary mode; ~6% in comfort-primary mode vs conventional on/off control, per Xi'an University of Architecture and Technology (2022).
MPC Demand Response Cost Function
Google's demand response patent minimizes three co-equal objectives simultaneously: energy consumption, occupant discomfort, and energy rate variability.
Hierarchical & Distributed MPC Architectures
Centralized optimization across large numbers of zones and subsystems is computationally intractable. The patent corpus converges on hierarchical decomposition as the dominant solution strategy for commercial-scale buildings.
Two-Level Hierarchical Decomposition
Johnson Controls Technology Company's 2018 patent describes a high-level MPC that generates an optimal load profile for each airside subsystem minimizing total energy cost, while a plurality of low-level airside MPCs each consume the assigned load profile and independently determine optimal zone-level temperature setpoints. This decomposition limits dimensionality and allows parallel computation.
Thermal Capacitance-Based Targeting
Johnson Controls' 2021 multi-level MPC patent treats the building's thermal mass as a controllable energy buffer — generating energy targets using the thermal capacitance of building spaces, then generating HVAC equipment setpoints consistent with those targets. This allows the system to exploit inherent thermal inertia to shift load away from peak electricity price periods.
Occupancy-Aware, Demand Response & Multi-Objective Control
Smart building MPC must balance energy cost minimization against occupant thermal comfort and respond to grid-level demand signals. The patent data reveals several converging strategies.
Fine-Grained Occupancy Prediction
Bosch's approach employs neural network predictors — including sequential and contextual architectures — to forecast occupant counts across multiple zones and future time slots. Critically, it explicitly accounts for occupancy misprediction by maintaining a misprediction-type distribution characterizing true negatives, false positives, false negatives, and true positives, then computing a total misprediction cost expectation. HVAC power per zone is set to optimize comfort weighted by predicted occupancy while minimizing misprediction costs.
Stochastic prediction robustnessHumidity-Inclusive Comfort Optimization
This patent extends the conventional temperature-only MPC formulation by simultaneously tracking both a desired temperature setpoint and a target humidity ratio for the building zone, with supply air temperature and humidity as jointly optimized decision variables. This formulation captures the full HVAC energy consumption associated with dehumidification, which is often the dominant cooling load in humid climates.
Dominant load in humid climatesDemand Response Event Handling
Google's patent implements DR control through a networked smart thermostat that minimizes a cost function comprising: (1) total energy consumption during the DR event, (2) a metric of occupant discomfort, and (3) the variability of energy consumption rates over the DR event duration. This formulation captures both energy cost and grid-friendliness objectives simultaneously — something reactive controllers cannot achieve.
Grid-integrated optimizationMulti-System Coordinated Optimization
NTU's two-stage MPC framework first optimizes a lighting and shading system using a visual comfort prediction model, then uses the predicted visual and thermal state as inputs to a second optimization stage governing the HVAC system. This coordinated approach reduces the total building energy footprint by exploiting the coupling between solar gain, daylighting, and thermal loads — recovering savings inaccessible to temperature-only control loops.
Solar–thermal couplingKey Players and Innovation Strategies
The MPC HVAC patent field is highly concentrated. A small number of vertically integrated industrial players hold the dominant portfolio positions, supplemented by academic institutions and software-first startups.
| Assignee | Filing Volume | Technical Focus | Key Jurisdictions | Strategic Approach |
|---|---|---|---|---|
| Johnson Controls Technology Co. / Tyco IP Holdings Most Prolific | 15+ discrete filings | Smart thermostatsHierarchical MPCCascaded models | US, WO | Full MPC stack ownership: sensor integration → optimization → equipment actuation |
| Robert Bosch GmbH | Multiple filings | Stochastic occupancy MPCNeural network predictors | US, CN, DE | Explicitly modeling prediction uncertainty as a cost term — technically distinctive |
| Honeywell International Inc. | Multiple active patents | Dynamic ventilationDistributed cost optimization | US, WO | Air quality and ventilation efficiency as primary energy levers |
| Nanyang Technological University | Multiple filings | Multi-objective MPCLighting & shading integration | WO, US | Research-to-patent pipeline: integrated air conditioning, lighting, and shading |
| Palo Alto Research Center (PARC) | 2 significant recent patents | Adaptive MHE model IDScalable embedded MPC | US | Addressing deployment barriers: computational cost and model maintenance |
| BrainBox AI Inc. | Patent family (WO, CA, KR, US) | Simulation-driven module selectionAI-driven control | WO, CA, KR, US | Software-first, AI-driven approach; no hand-crafted physics models |
Track MPC HVAC competitors in real time
Set up automated patent monitoring for any assignee in this landscape with PatSnap Analytics.
Key Takeaways from the MPC HVAC Patent Corpus
MPC fundamentally outperforms reactive control by optimizing over a receding time horizon. As quantified by Xi'an University of Architecture and Technology, MPC achieves up to 19.4% energy savings versus conventional on/off control while keeping PMV indices within acceptable comfort ranges.
Hierarchical decomposition is the dominant architecture for scalability. Johnson Controls' distributed airside optimization demonstrates that separating high-level energy target generation from low-level setpoint optimization allows MPC to scale to large multi-zone commercial buildings without combinatorial explosion.
Thermal model adaptability is critical for sustained performance. ABB AG's probabilistic diagnosis and targeted model updates prevent energy and comfort degradation as building conditions evolve over time — a limitation of static model approaches.
Occupancy prediction quality directly gates energy savings potential. Robert Bosch GmbH's explicit accounting for misprediction costs — rather than treating forecasts as deterministic — produces control policies robust to stochastic false positive and negative events, avoiding wasteful over-conditioning and occupant discomfort.
Computational tractability remains an active research frontier. Palo Alto Research Center's 2024 patent addresses MPC deployment on embedded hardware through structured matrix factorization, indicating the field is actively transitioning from cloud-dependent to edge-deployable implementations. For teams monitoring this transition, PatSnap's platform tracks filing velocity across all relevant assignees. The International Energy Agency has identified smart building controls as a key pathway to meeting global net-zero targets, and ASHRAE standards increasingly reflect MPC-compatible performance metrics.
The MPC HVAC Optimization Workflow
From sensor data ingestion to equipment actuation, the MPC receding-horizon loop integrates thermal modeling, occupancy prediction, and multi-objective optimization at every time step.
Model Predictive Control for HVAC — key questions answered
Research from Xi'an University of Architecture and Technology reports that MPC achieves energy savings of up to 19.4% versus conventional on/off control while maintaining the PMV index within an acceptable comfort range, with the system described as saving approximately 6% energy when comfort is the primary objective.
The receding-horizon strategy involves a thermal model predicting future zone temperatures over a defined horizon, an optimizer minimizing a cost function subject to comfort and equipment constraints, and only the first control action being applied before the process repeats. This allows MPC to preemptively counteract anticipated thermal loads rather than reacting to them after the fact.
A high-level MPC generates an optimal load profile for each airside subsystem that minimizes total energy cost, while a plurality of low-level airside MPCs each consume the assigned load profile and independently determine optimal zone-level temperature setpoints. This decomposition limits the dimensionality of each optimization problem and allows parallel computation.
Robert Bosch GmbH's approach explicitly accounts for occupancy misprediction by maintaining a misprediction-type distribution (characterizing true negatives, false positives, false negatives, and true positives) and computing a total misprediction cost expectation. HVAC power per zone is then set to optimize occupant thermal comfort weighted by predicted occupancy while minimizing misprediction costs, making the control policy robust to stochastic prediction errors.
Yes. Google's patent on controlling an HVAC system during demand response events implements DR control through a networked smart thermostat that minimizes a cost function comprising: (1) total energy consumption during the DR event, (2) a metric of occupant discomfort, and (3) the variability of energy consumption rates over the DR event duration. This formulation captures both energy cost and grid-friendliness objectives.
The University of Florida Research Foundation's MPC formulation simultaneously tracks both a desired temperature setpoint and a target humidity ratio for the building zone, with supply air temperature and humidity as jointly optimized decision variables. This formulation captures the full HVAC energy consumption associated with dehumidification, which is often the dominant cooling load in humid climates.
Still have questions about MPC for HVAC? Let PatSnap Eureka answer them for you.
Ask PatSnap Eureka an MPC QuestionAccelerate Your Smart Building R&D with AI-Powered Patent Intelligence
Join 18,000+ innovators already using PatSnap Eureka to accelerate their R&D. Search, analyse, and monitor the full MPC HVAC patent landscape — from thermal modeling to demand response integration.
References
- Building HVAC System with Modular Cascaded Model — Johnson Controls Technology Company, 2021
- System and Method for Modeling, Parameter Estimation and Adaptive Control of Building HVAC System — Palo Alto Research Center Incorporated, 2023
- A Building Thermal Comfort Model Predictive Control Method and System — Xi'an University of Architecture and Technology, 2022
- Adaptive Modeling Method and System for MPC-Based Building Energy Control — ABB AG, 2019
- Method and System for Scalable Embedded Model Predictive Control of HVAC Systems — Palo Alto Research Center Incorporated, 2024
- HVAC System Using Model Predictive Control with Distributed Low-Level Airside Optimization — Johnson Controls Technology Company, 2018
- HVAC System Using Model Predictive Control with Distributed Low-Level Airside Optimization and Airside Power Consumption Model — Johnson Controls Technology Company, 2018
- Building HVAC System with Multi-Level Model Predictive Control — Johnson Controls Technology Company, 2021
- Building HVAC System with Multi-Level Model Predictive Control — Johnson Controls Tyco IP Holdings LLP, 2023
- Distributed HVAC System Cost Optimization — Honeywell International Inc., 2015
- Building System with Multi-Tiered Model Based Optimization for Ventilation and Setpoint Control — Johnson Controls Tyco IP Holdings LLP, 2022
- HVAC Control Fine-Grained Occupancy Pattern Estimation — Robert Bosch GmbH, 2023
- Model Predictive Control-Based Building Climate Controller Incorporating Humidity — University of Florida Research Foundation, 2021
- Method and Control System for Controlling Building Service Systems — Nanyang Technological University, 2024
- Method and Control System for Controlling an Air-Conditioning System — Nanyang Technological University, 2023
- Controlling an HVAC System During Demand Response Events — Google Inc., 2021
- A Method for Creating Demand Response Determination Model for HVAC System — Seokyoung Systems, 2021
- Dynamic Ventilation Control for a Building — Honeywell International Inc., 2022
- HVAC System with Building Infection Control and Sustainability and Emissions Controls — Johnson Controls Tyco IP Holdings LLP, 2023
- System and Method for Dynamic Control of HVAC Components of a Building — BrainBox AI Inc., 2023
- U.S. Department of Energy — Building Energy Efficiency
- International Energy Agency — Smart Buildings and Net-Zero
- ASHRAE — Building Performance Standards and MPC-Compatible Metrics
All data and statistics on this page are sourced from the references above and from PatSnap's proprietary innovation intelligence platform. Patent analysis conducted via PatSnap Eureka across 60+ documents spanning US, EU, WO, JP, KR, CN, AU, and DE jurisdictions, filed 2012–2026.
PatSnap Eureka searches patents and research to answer instantly.