Thermal Network Modeling With Sub-Library Regression Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for generating thermal network models face challenges in creating appropriate physical phenomenon models due to increased search spaces and unstable learning, particularly in diverse heat transfer phenomena.
Innovation Solution
An information processing device that utilizes a library storage unit to store sub-libraries of nonlinear basis functions, a detection result acquisition module, a regression equation generation module, a correction module, and an output control module to generate and refine linear regression equations for temperature prediction by sparse estimation and hyperparameter correction, using sub-libraries for heat conduction, radiation, convection, and heat generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional techniques are used to generate thermal network models, then the search space increases and learning becomes unstable, but the model generation capability is maintained
Solution Approach 1:
The patent segments the thermal network model into multiple sub-libraries, each containing basis functions for specific heat transfer mechanisms (conduction, convection, radiation). This segmentation reduces the overall search space by organizing functions thematically rather than treating all possible thermal phenomena as a single large space, thereby improving learning stability while maintaining comprehensive modeling capability.
Solution Approach 2:
The patent changes the parameter representation by using basis functions with physical meanings (thermal conductivity, heat capacity, convection coefficients) as parameters. This transforms the learning problem from searching through arbitrary mathematical functions to selecting and combining physically meaningful parameters, reducing search space complexity and stabilizing learning while preserving physical accuracy.
2Adaptability or versatility
If diverse heat transfer phenomena are modeled, then the model comprehensiveness improves, but the search space increases and learning becomes unstable
Solution Approach 1:
The patent creates a universal thermal network model framework that can handle diverse heat transfer phenomena through a common structure of sub-libraries. Each sub-library (conduction, convection, radiation) follows the same functional interface, allowing the system to universally apply the same modeling approach to different physical phenomena without requiring separate learning processes for each, thus maintaining comprehensiveness while ensuring learning stability.
Solution Approach 2:
By segmenting diverse heat transfer phenomena into distinct sub-libraries with dedicated basis functions, the patent enables the system to handle multiple physical phenomena simultaneously without creating a monolithic search space. This segmentation allows independent optimization and stable learning within each sub-domain while maintaining overall model versatility.
3Reliability
If the search space is reduced, then learning stability improves, but the model accuracy may deteriorate
Solution Approach 1:
The patent changes the parameter representation to use physically meaningful basis functions (thermal conductivity, heat capacity, convection coefficients) that directly relate to physical phenomena. This ensures that even within a reduced search space, the model maintains high accuracy by searching through physically relevant parameters rather than arbitrary mathematical functions, thus resolving the contradiction between search space size and model accuracy.
Solution Approach 2:
The patent composites different heat transfer mechanisms (conduction, convection, radiation) into a unified thermal network model through the sub-library structure. This composite approach allows the model to accurately represent complex thermal phenomena by combining multiple physical mechanisms while maintaining a manageable search space through systematic organization of the composite components.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
According an embodiment, an information processing device (1) includes a storage unit (10) storing therein library information, a detection result acquisition module (111), a regression equation generation module (112), a correction module (113), a regression equation regeneration module (114), and an output control module (115). The correction module (113) corrects generation probabilities and a hyperparameter of machine learning on the basis of loss functions of linear regression equations. The regression equation regeneration module (114) extracts nonlinear basis functions on the basis of the corrected generation probabilities from sub-libraries including the nonlinear basis functions, generates a plurality of linear regression equations obtained by combining the nonlinear basis functions of the plurality of types of sub-libraries, estimates coefficients of the linear regression equations by machine learning using the corrected hyperparameter, and calculates loss functions of the linear regression equations.