Adaptive model predictive control of building HVAC using moving horizon estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing HVAC control systems face challenges in accurately modeling building environments due to heterogeneity of components, uncertainty in data, and complexity, limiting the deployment of model predictive control (MPC) for energy efficiency and occupant comfort.
Innovation Solution
A physics-constrained, data-driven model using a differentiable physics model and neural networks is combined with a moving horizon estimation (MHE) algorithm to identify and update HVAC system dynamics, enabling adaptive model predictive control (MPC) that learns from past state values and inputs, and applies current inputs to controlled components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional HVAC control systems are used, then device complexity is reduced, but manufacturing precision and measurement precision deteriorate due to inability to accurately model building environments
Solution Approach 1:
The patent merges physics-based models with data-driven neural network models into a hybrid architecture. The physics model provides thermodynamic relationships and constraints, while the neural network learns complex nonlinear behaviors from operational data. This combination achieves accurate building environment modeling without requiring complete first-principles knowledge of all system components.
Solution Approach 2:
The patent introduces moving horizon estimation (MHE) as an intermediary layer between measurements and model parameters. MHE recursively estimates system states and identifies model parameters by optimizing a cost function over a moving time horizon, enabling accurate parameter identification even with noisy or incomplete measurements.
2Measurement precision
If detailed building models are used, then manufacturing precision improves, but device complexity increases making deployment difficult
Solution Approach 1:
The patent segments the building model into multiple thermal zones, each with its own thermal mass and temperature dynamics. This segmentation allows the complex building to be modeled as interconnected simpler subsystems, making the overall model more manageable while maintaining accuracy for control purposes.
Solution Approach 2:
The patent transforms the modeling approach from requiring detailed physical parameters of all components to using a reduced set of effective thermal parameters that capture dominant heat transfer behaviors. The neural network learns these effective parameters from operational data, avoiding the need for comprehensive building audits and detailed component specifications.
3Productivity
If adaptive control with learning is implemented, then productivity improves through energy savings, but device complexity increases due to online parameter updates
Solution Approach 1:
The patent implements continuous feedback loops where sensor measurements of actual building temperatures and HVAC performance are fed back to the controller. The moving horizon estimation recursively updates model parameters based on this feedback, and the model predictive control adjusts setpoints and control actions to minimize energy consumption while maintaining comfort constraints.
Solution Approach 2:
The patent transitions from static HVAC control to dynamic adaptive control where model parameters and control strategies evolve over time. The neural network adapts to changing building usage patterns, weather conditions, and system degradation, while the MHE continuously identifies updated parameters online, enabling the system to optimize performance under varying operating conditions.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The proposed method achieves substantial energy savings and maintains occupant comfort by accurately predicting HVAC system behavior, outperforming traditional control methods with an 18.08% overall energy savings and matching the performance of exact MPC models without requiring detailed building models.
Implementation Method 1
a differentiable physics model of the building is determined. The differentiable physics model defines thermodynamic relationships between zones of the building and a heating, ventilation, and air-conditioning (HVAC) system
Data Source
AI summary
A differentiable physics model of a building is used that defines thermodynamic relationships between zones of the building and a heating, ventilation, and air-conditioning (HVAC) system. A physics-constrained, data driven model learns behaviors of controlled components of the HVAC system. For each of a series of times during online operation of the HVAC system, past state values are recorded representing a performance of the HVAC system in the building and past inputs to the HVAC system to maintain the states. The past state values and the past inputs are input into the differentiable physics model and the data driven model to: jointly update first parameters of the differentiable physics model and second parameters of the data driven model, e.g., using moving horizon estimation; and determine a current input to the controlled components, e.g., using model predictive control.