Multi-Objective Optimization With Reliability-Based Penalty Terms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-objective optimization methods using characteristic prediction models struggle to accurately optimize values of explanatory variables due to unreliable training data, leading to inaccurate prediction values and suboptimal solutions.
Innovation Solution
Perform first multi-objective optimization to generate a solution set considering reliability, using a model to predict values and calculate an index value indicating reliability, then perform second multi-objective optimization with a penalty term based on the specified index value to enhance solution accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multi-objective optimization is performed using a characteristic prediction model trained with unreliable training data, then the optimization process can be executed, but the prediction values become inaccurate and the solutions are suboptimal
Solution Approach 1:
The patent performs preliminary multi-objective optimization for each characteristic variable individually before performing the final multi-objective optimization. This preliminary action generates solution sets that consider reliability indices for each variable separately, allowing the system to prepare optimized solutions with reliability assessments in advance, thereby improving the accuracy of the final optimization results even when training data is unreliable
Solution Approach 2:
The patent introduces a feedback mechanism where reliability indices are calculated for each characteristic variable based on the solution sets from preliminary optimization. These reliability indices are then fed back into the final multi-objective optimization process as constraints or weighting factors, allowing the system to iteratively improve solution accuracy by considering the reliability of predictions for each variable
2Reliability
If extensive model training is performed to improve prediction accuracy, then the reliability of prediction values increases, but the time and computational resources required increase significantly
Solution Approach 1:
The patent segments the optimization process into multiple independent multi-objective optimization tasks, one for each characteristic variable. Instead of training a single comprehensive model extensively, the system performs separate optimizations for each variable using existing prediction models, thereby achieving reliable results without requiring extensive additional training time for a unified model
Solution Approach 2:
The patent enables the optimization system to self-service by performing multi-objective optimization directly on existing prediction models without requiring external retraining. The system uses the prediction models as-is and applies optimization algorithms to generate reliable solutions, eliminating the need for time-consuming model retraining while still achieving high reliability through the optimization process itself
Data Source
AI summary
A storage medium storing an information processing program that causes a computer to execute a process that includes generating a solution set of a combination of the value and the index value by performing first multi-objective optimization by using a first objective function that searches for a value of the characteristic variable and a second objective function that searches for an index value that indicates reliability of the value; specifying an index value included in a combination that serves as a solution in a case where a characteristic variable is a certain value in the generated solution set; and generating a solution set of a combination of the respective values of the plurality of characteristic variables by performing second multi-objective optimization by using an objective function that searches for a value of each of a plurality of characteristic variables.