Multi-Acquisition Bayesian Optimization for High-Cost Digital Twins
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Bayesian optimization techniques for optimizing computational models in industrial processes are inefficient, sensitive to bias, and data-inefficient due to the disposal of data from each execution, especially when dealing with high-cost computational objective functions.
Innovation Solution
A data-efficient multi-acquisition strategy that employs multiple Bayesian acquisition functions in parallel, leveraging proper orthogonal decomposition to generate a surrogate model and utilize multi-fidelity Gaussian process regression, combining exploratory and exploitative approaches to optimize input parameters more efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If Bayesian optimization is used to optimize computational models, then optimization capability is improved, but sensitivity to algorithm configuration and data inefficiency worsen
Solution Approach 1:
The patent segments the single Bayesian optimization process into multiple parallel acquisition functions (e.g., expected improvement, probability of improvement, upper confidence bound). Each acquisition function operates independently with different exploration-exploitation characteristics, reducing sensitivity to any single configuration while maintaining overall optimization effectiveness.
Solution Approach 2:
The patent implements a multi-acquisition function framework where multiple acquisition functions serve different purposes (exploration, exploitation, uncertainty reduction) within a single optimization system. This multi-functional approach allows the system to adapt to different problem characteristics without requiring separate optimization processes.
2Device complexity
If single acquisition function is used in Bayesian optimization, then algorithm simplicity is maintained, but bias and data efficiency worsen
Solution Approach 1:
The patent merges multiple acquisition functions into a unified framework that processes data from all functions simultaneously. The multi-acquisition approach combines information from diverse exploration and exploitation strategies, preventing information loss that occurs when using a single acquisition function that must discard data not aligned with its specific objective.
3Measurement precision
If high-cost computational objective functions are evaluated frequently, then optimization accuracy is improved, but computational cost and time increase
Solution Approach 1:
The patent applies partial action by using surrogate models to approximate the high-cost computational objective function. The surrogate model provides inexpensive evaluations that guide the optimization process, requiring only selective and partial evaluations of the expensive true objective function, thereby reducing computational time while maintaining optimization accuracy.
Solution Approach 2:
The patent introduces surrogate models as intermediary representations between the optimization algorithm and the high-cost computational objective function. These surrogate models act as mediators that provide approximate evaluations at low computational cost, reducing the frequency and impact of expensive true function evaluations while preserving optimization effectiveness.
Data Source
AI summary
A method of optimizing parameters for an industrial process is described, along with media and systems, using a digital twin, physics based model and multiple types of acquisition functions. Output data from the model is analyzed by multiple types of Bayesian acquisition functions, such as an expected improvement acquisition function and a model variance acquisition function. The different acquisition functions tune better parameters, and then the model is re-run in parallel for each to output more data. The data from one acquisition function's run of the model may be co-mingled with data from the other acquisition function's run of the model such that the acquisition functions' exploration and exploitation of the parameter space are intertwined, thus achieving a more globally optimal solution than using just one type of Bayesian acquisition function.