Deep learning-based cooling system temperature prediction apparatus according to physical causality and method therefor
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cooling system temperature prediction methods using artificial neural networks fail to accurately analyze the physical relationship between inputs and outputs, making it difficult to understand the prediction process and trace errors, especially in systems with low data variability.
Innovation Solution
A deep learning-based cooling system temperature prediction apparatus and method that uses multiple artificial neural network submodels with input layers, hidden layers, and output layers to predict battery and motor temperatures by reflecting physical causality, incorporating control variables, environment variables, and time variables, and utilizing feedback variables for learning and re-learning processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional statistical analysis is used to determine input/output variables for neural network temperature prediction, then the prediction process is simple to implement, but the physical relationship between inputs and outputs cannot be analyzed
Solution Approach 1:
The patent segments the temperature prediction task into multiple submodels, each responsible for predicting specific intermediate physical quantities (battery pump flow speed, cooling water flow rates, inlet temperatures) before final temperature prediction. This segmentation allows each submodel to focus on a specific physical relationship, making the overall system both implementable and physically interpretable.
Solution Approach 2:
The patent introduces intermediate physical quantities as mediator variables between input control variables and output temperatures. These intermediaries (flow speeds, inlet temperatures) represent actual physical states in the cooling system, enabling traceable physical analysis while maintaining systematic prediction capability.
2Device complexity
If a single artificial neural network model is used for temperature prediction, then the model structure is simple, but it is difficult to trace the prediction process and identify errors
Solution Approach 1:
The patent divides the single neural network model into multiple sequential submodels, where each submodel performs a specific prediction task. This segmentation creates a traceable prediction pipeline where errors can be identified at specific stages, while the overall model structure remains manageable through modular organization.
Solution Approach 2:
The patent transforms the prediction process from a single-dimensional black-box model to a multi-dimensional transparent structure by introducing intermediate physical quantities as additional prediction dimensions. This allows error tracing across multiple prediction stages while maintaining systematic organization.
3Measurement precision
If multiple submodels are used to predict temperatures by reflecting physical causality, then prediction accuracy and physical analysis capability are improved, but the system complexity increases
Solution Approach 1:
The patent segments the complex temperature prediction system into specialized submodels, each handling specific physical relationships. This segmentation improves prediction accuracy for each component while organizing system complexity into manageable modular units with clear functional boundaries.
Solution Approach 2:
The patent changes the prediction parameters from direct temperature prediction to multi-stage prediction of intermediate physical quantities (flow speeds, temperatures) followed by final temperature prediction. This parameter transformation enables more accurate physical modeling while structuring complexity through systematic parameter progression.
Data Source
AI summary
A deep learning-based cooling system temperature prediction apparatus has an artificial neural network modeled by connecting a plurality of artificial neural network submodels each including an input layer, a hidden layer, and an output layer is used. A pump flow speed, a cooling water flow rate, a battery inlet cooling water temperature, a motor inlet cooling water temperature, a radiator outlet cooling water temperature, a battery temperature, and a motor temperature are predicted by inputting at least one of a predetermined control variable, an environment variable, or a time variable to the plurality of artificial neural network submodels in accordance with a physical causality. A number of the plurality of artificial neural network submodels and the control variables or environment variables that are sequentially input to each submodel depend on divisional control and integral control of the cooling system.