Hybrid AI Modeling with Auto-ML for Dynamic System Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional vehicle dynamics models, whether 'white-box' or 'black-box', face challenges in accuracy and computational efficiency, with hybrid models being computationally expensive and requiring significant memory, while hybrid AI algorithms can combine physical and data-driven approaches but struggle with interpretability and adaptability.
Innovation Solution
A computer-implemented method and apparatus for generating a hybrid artificial intelligence algorithm that uses a system state vector combining analytical/statistical equations with AI algorithms, employing Auto-ML methods to select and optimize neural networks and differential equations for dynamic system modeling, allowing for accurate and efficient description of physical system behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If hybrid ML models use ensemble methods with multiple ML models to solve prediction problems, then model accuracy is improved, but computational cost increases significantly
Solution Approach 1:
The patent segments the prediction problem into two distinct parts: a physics-based component that handles conservation laws and constraints, and a machine learning component that handles complex nonlinear relationships. This segmentation allows each component to be optimized independently, reducing the overall computational burden compared to using multiple full ML models in an ensemble.
Solution Approach 2:
The patent introduces a hybrid model structure that acts as an intermediary between pure physics-based models and pure ML models. The physics-based component serves as a mediator that incorporates domain knowledge and constraints, while the ML component learns residual patterns, achieving high accuracy without requiring multiple complete ML models.
2Measurement precision
If vehicle dynamics models increase complexity to improve accuracy, then model precision is improved, but implementation cost and optimization time increase
Solution Approach 1:
The patent changes the parameters of the model by incorporating physics-based constraints and conservation laws as fixed structural elements, while only requiring optimization of fewer ML model parameters. This reduces the overall model complexity compared to fully parameterized ML models while maintaining or improving accuracy through the physics-guided structure.
3Reliability
If Auto-ML methods search larger model spaces to find optimal models, then model performance is improved, but development time increases
Solution Approach 1:
The patent performs preliminary action by pre-defining the physics-based component structure with conservation laws and constraints before the ML model search begins. This preliminary structuring guides the Auto-ML search process, constraining it to search within a physically meaningful subspace, which reduces the effective search space and development time while maintaining model performance.
Data Source
AI summary
A computer-implemented method for generating a hybrid artificial intelligence algorithm. The method includes: providing a system state vector that includes at least two system states of the physical system, wherein at least one of the at least two states can be calculated by at least one analytical and/or statistical equation, and wherein at least one other of the at least two states can be calculated by at least one artificial intelligence algorithm to be ascertained; carrying out at least one automatic learning method for ascertaining a plurality of artificial intelligence algorithms using which at least the other of the at least two states can be calculated in each case; and selecting at least one artificial intelligence algorithm from the plurality of artificial intelligence algorithms as a function of at least one selection criterion to generate the hybrid artificial intelligence algorithm coupled with the at least one equation.