DAG Proxy Modeling for Gray-Box Optimization Input Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gray box optimization methods are difficult to apply in simulations where intermediate outputs can be operated in any manner, making it challenging to estimate inputs for desired outputs in systems with multiple connected functions.
Innovation Solution
An optimization support device and method that utilizes a directed acyclic graph to generate proxy models for each function, determine input ranges based on parent node outputs, select input points within those ranges, and update models using sampled data to calculate input points for desired outputs of objective functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If gray box optimization is applied to systems with intermediate outputs that can be operated in any manner, then optimization capability is improved, but applicability and accuracy deteriorate because existing methods cannot handle freely operable intermediate outputs
Solution Approach 1:
The patent segments the objective function into multiple functions connected by a directed acyclic graph, where each function has its own proxy model. This segmentation allows the system to handle intermediate outputs individually, tracking their feasible ranges separately, which enables proper handling of freely operable intermediate outputs while maintaining optimization capability.
Solution Approach 2:
The patent changes the parameter representation by introducing prediction distributions for intermediate outputs and determining feasible input ranges based on these distributions. Instead of treating intermediate outputs as fixed values, the system uses probabilistic parameters (mean and standard deviation) to represent them, enabling accurate handling of freely operable intermediate outputs.
2Adaptability or versatility
If the search space for intermediate outputs is not constrained, then flexibility is improved, but computational efficiency deteriorates due to unnecessary evaluations
Solution Approach 1:
The patent performs preliminary determination of feasible input ranges for intermediate outputs before conducting the full optimization search. By using proxy models to predict output ranges in advance, the system pre-constrains the search space, eliminating unnecessary evaluations while preserving flexibility in how intermediate outputs are operated.
Solution Approach 2:
The patent applies partial constriction to the search space by determining feasible ranges for intermediate outputs based on parent node predictions. Instead of constraining the entire search space, only the relevant intermediate output ranges are partially constrained, maintaining flexibility where needed while improving efficiency in critical areas.
3Measurement precision
If proxy models are updated with sampled data, then prediction accuracy is improved, but the number of required samples increases computational cost
Solution Approach 1:
The patent applies local quality by updating proxy models selectively based on the specific functions and intermediate outputs being optimized. Instead of uniformly sampling and updating all proxy models, the system focuses computational resources on local areas where prediction accuracy most impacts the final optimization result, reducing overall computational cost while maintaining necessary accuracy.
Data Source
AI summary
An optimization support device for calculating an input point according to a desired output of an objective function, wherein the objective function is a model in which relationships between functions are represented by a directed acyclic graph with functions as nodes and inputs/outputs as edges. The optimization support device includes processors configured to generate proxy models indicating prediction distributions of output values for each function, determine an input range of a child node function from an output range of a parent node function estimated based on the proxy model, select an input point within the determined input range, update the proxy model using sampled data for the selected input point, and calculate the input point according to the desired output using a prediction distribution of objective function output values calculated based on the proxy model thereby supporting decision making for optimization problems.