Live Multi-Objective Optimization with Irregular Operation Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large-scale value-based non-linear models with numerous decision variables and constraints face difficulties in feasible search spaces, leading to suboptimal implementations of multi-objective optimizations due to improper implementation or uncontrollable perturbations, resulting in irregular operations.
Innovation Solution
A live optimization system combining Multi-Objective Evolutionary Algorithms (MOEAs) with an Irregular Operations (IROP) system, which performs initial optimizations, tracks metrics, and detects anomalies by comparing live data to expected progressions, initiating re-optimizations with reduced decision variables and objectives to address irregularities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If large-scale multi-objective evolutionary algorithms are used to optimize problems with numerous decision variables and constraints, then optimization coverage and solution quality improve, but computational complexity and resource requirements increase significantly
Solution Approach 1:
The patent segments the large-scale optimization problem into multiple smaller sub-problems by dividing the set of decision variables into groups. Each sub-problem is optimized independently with fewer variables, reducing computational complexity while maintaining overall optimization quality. The segmented solutions are then integrated to form the complete optimization result.
Solution Approach 2:
The patent implements dynamic adjustment of optimization parameters and problem segmentation based on live operational data. When irregular operations are detected, the system dynamically re-optimizes only the affected subsets of variables rather than re-running the complete optimization, adapting the computational effort to actual system needs.
2Reliability
If complete multi-objective optimization is performed initially, then comprehensive solution coverage is achieved, but implementation time and computational resource consumption increase
Solution Approach 1:
The patent performs preliminary complete multi-objective optimization to establish baseline optimal solutions and expected operational progressions before actual implementation. This preliminary optimization creates reference trajectories for key metrics that enable later detection of irregular operations without requiring repeated full optimizations.
Solution Approach 2:
The system implements continuous feedback monitoring of live operational data against the expected progressions established by preliminary optimization. When deviations indicating irregular operations are detected, targeted re-optimization is triggered only for affected variable subsets, reducing implementation time while maintaining solution reliability.
3Manufacturing precision
If the optimization system accounts for all constraints and variables, then feasible solution discovery becomes possible, but the search space becomes vastly infeasible and computation becomes impractical
Solution Approach 1:
The patent divides the complete set of constraints and variables into multiple subsets, creating several smaller optimization problems. Each subset can be solved independently and more efficiently, while the collection of subset solutions collectively satisfies all original constraints. This segmentation makes the optimization computationally tractable.
Solution Approach 2:
The system performs optimization on partial sets of variables and constraints at different times rather than attempting to optimize everything simultaneously. By focusing computational effort on relevant subsets based on current operational context and irregular operation detection, the system achieves adequate constraint satisfaction without exhaustive computation.
Data Source
AI summary
Systems and methods are provided for generating an initial optimized baseline solution to a multi-objective problem. As the baseline solution is implemented, live (e.g., real-time or near real-time) data associated with one or more parameters may be received and compared to expectations of those parameters with the implementation of the initial optimized solution. If a deviation is detected between the expectation of the time progression of the parameters and live data associated with the parameter, then that deviation may be compared to a threshold. If the deviation meets a threshold condition, then an irregular operation may be declared and a new baseline solution may be implemented. The new baseline solution may be obtained as a re-optimized solution.