Evolutionary Algorithm Variable Segmentation for High-Dimensional Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for multi-objective optimization with a large number of variables are inefficient, often requiring extensive a priori knowledge or simplifying the problem, which is costly or not feasible, especially in complex scenarios like satellite clusters or airline networks, where variable interactions are highly coupled.
Innovation Solution
The introduction of additional Boolean variables, referred to as flags, which allow for selective consideration of original variables during evolutionary algorithm iterations, enabling the identification of most significant variables and reducing the search space while maintaining diversity of solutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the number of variables in multi-objective optimization problems increases, then the problem can model more complex scenarios (e.g., satellite clusters, airline networks), but current state-of-the-art methods become incapable of handling the problem efficiently
Solution Approach 1:
The patent segments the large set of variables into multiple subsets or groups. Instead of optimizing all variables simultaneously, the algorithm divides the search space into manageable segments that can be processed independently or in parallel, enabling efficient handling of high-dimensional optimization problems while maintaining the ability to model complex scenarios.
Solution Approach 2:
The patent transforms the high-dimensional optimization problem by introducing an additional dimension through hierarchical structuring. Variables are organized in a hierarchy where higher-level variables control groups of lower-level variables, effectively reducing the computational complexity by adding a structural dimension to the problem representation.
2Productivity
If variables are reduced through sensitivity analysis, then existing methods can be used, but extensive a priori knowledge is required which is costly to attain
Solution Approach 1:
The patent performs preliminary structuring of variables into hierarchical groups before the optimization process begins. This preliminary organization enables the algorithm to efficiently identify and focus on critical variable groups without requiring extensive a priori sensitivity analysis, thus achieving computational tractability while minimizing the time needed for problem preparation.
Solution Approach 2:
The optimization algorithm automatically identifies and focuses on the most influential variable groups during the optimization process itself, without requiring external sensitivity analysis. The hierarchical structure enables the algorithm to self-organize and prioritize variables based on their impact, eliminating the need for costly a priori knowledge acquisition.
3Productivity
If variables are reduced to make the problem tractable, then computational resources are saved, but the variable interactions may be so highly coupled that reduction is not possible
Solution Approach 1:
The patent segments highly coupled variables into interconnected hierarchical groups rather than treating them as a monolithic block. Each group maintains its internal couplings while being managed as a unified unit in the hierarchical structure, allowing computational efficiency through structured processing while preserving the complex interaction patterns within each segment.
Solution Approach 2:
The patent resolves the complexity of highly coupled variables by introducing a hierarchical dimension. Variables that are highly coupled in the original space are organized into hierarchical groups where their interactions are captured at higher levels of the hierarchy, transforming an intractable flat optimization problem into a manageable hierarchical optimization problem.
Data Source
AI summary
Systems and methods are provided to enable vector scalability in evolutionary algorithms to enable execution of optimization problems having a relatively large number of variables. A subset of the total number of variables of a chromosome data structure may be considered relative to a baseline known solution for the purpose of evaluating one or more objective functions of the evolutionary algorithm.