A prediction method based on fractional order displacement for dynamic multiobjective optimization.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Dynamic multiobjective optimization problems (DMOPs) are increasingly solved using prediction-based methods.
• Existing methods often use limited historical data (2-3 previous environments), potentially reducing prediction accuracy.
• Utilizing all historical data is computationally expensive and may not improve predictions due to weak correlations.

Purpose of the Study:

• To propose a novel prediction method, Fractional Displacement-based Prediction Method (FDPM), for DMOPs.
• To enhance prediction accuracy while managing computational costs in dynamic optimization.
• To identify and utilize relevant historical optimal solutions for improved future predictions.

Main Methods:

• Developed a prediction model to identify historical optimal solutions with significant correlation to new solutions.
• Trained model parameters using a series of previous optimal solutions.
• Employed identified relevant historical solutions to predict optimal solutions in new environments.

Main Results:

• The proposed FDPM demonstrated superior performance compared to five state-of-the-art algorithms.
• Evaluated on fourteen benchmark problems with diverse characteristics.
• FDPM achieved a balance between prediction accuracy and computational efficiency.

Conclusions:

• FDPM offers a more effective approach for solving DMOPs by intelligently leveraging historical data.
• The method provides a practical solution for dynamic optimization challenges where prediction accuracy and efficiency are critical.
• Selective use of historical data in FDPM outperforms methods relying on limited or extensive history.