HVAC system using interconnected neural networks and online learning and operation method thereof
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing HVAC systems face challenges in optimizing energy efficiency due to the complexity of nonlinear relationships between HVAC loads, indoor temperature, and outdoor environment, requiring numerous unknown parameters and varying system types across different buildings.
Innovation Solution
An HVAC system utilizing interconnected artificial neural networks (ANNs) trained for respective subsystems, which includes air conditioning sensor units, an HVAC device, and a predictive controller to generate operational data and adjust input power based on environmental data, improving modeling accuracy and optimizing schedule generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If physics-based modeling is used to optimize HVAC system operation, then the optimization problem can be formulated using physical parameters and mechanical/thermodynamics equations, but numerous unknown parameters are required and the model cannot be applied universally to various buildings
Solution Approach 1:
The patent replaces the physics-based mechanical/thermodynamic modeling approach with a data-driven artificial neural network model. The ANN learns building thermal response characteristics directly from operational data without requiring explicit physical equations, thereby achieving both accurate modeling and universal applicability across different building types and HVAC systems.
Solution Approach 2:
Instead of using explicit physical equations to represent building thermal dynamics, the patent creates a computational copy of the building's thermal response behavior through the ANN model. The network learns and replicates the complex nonlinear relationships between HVAC inputs and indoor temperature responses from operational data, enabling accurate prediction without requiring underlying physical parameter knowledge.
2Reliability
If physics-based modeling is used to optimize HVAC system operation, then the optimization problem can be formulated using physical parameters, but numerous parameters are unknown and need to be extracted using sophisticated estimation techniques
Solution Approach 1:
The patent replaces the complex parameter extraction process inherent in physics-based modeling with a data-driven ANN approach. The neural network automatically learns building-specific thermal characteristics from operational data during training, eliminating the need for manual parameter measurement and estimation while maintaining modeling accuracy.
Solution Approach 2:
The ANN model performs self-learning and self-calibration by automatically extracting building thermal response characteristics from operational data during the training phase. This eliminates the need for external parameter estimation techniques and allows the model to adapt to each specific building's unique thermal properties without manual intervention.
3Reliability
If interconnected artificial neural networks are used to model HVAC subsystems, then modeling accuracy of building thermodynamics is improved, but the system complexity increases
Solution Approach 1:
The patent divides the building thermal system into distinct controllable subsystems (e.g., cooling system, heating system, ventilation system), each modeled by a separate ANN. This segmentation allows each network to focus on learning the specific thermal response characteristics of individual subsystems, improving overall modeling accuracy while keeping each individual network relatively simple and manageable.
Solution Approach 2:
The patent integrates multiple specialized ANN models into a unified predictive controller that combines the thermal response predictions from different subsystem models. This merging allows the system to capture complex interactions between multiple HVAC components and building thermal zones, achieving high overall modeling accuracy through the coordinated operation of interconnected networks.
Data Source
AI summary
The present disclosure provides a heating, ventilation, and air conditioning (HVAC) system including interconnected artificial neural networks trained for respective subsystems that are required for building temperature control. The HVAC system includes: an air conditioning sensor units installed in or outside a building to detect environmental data; an HVAC device configured to supply thermal energy into an inner space of the building using input power; and a predictive controller configured to generate operational data based on the environmental data and control the HVAC device by adjusting the input power.