Evolutionary Dynamic Multiobjective Optimization Via Kalman Filter Prediction.

Area of Science:

• Computational Intelligence
• Optimization Theory
• Evolutionary Computation

Background:

• Multiobjective evolutionary algorithms (MOEAs) excel at static optimization but are underutilized for dynamic problems.
• Dynamic optimization requires algorithms that can track shifting optima as objectives change over time.
• Existing research lacks sufficient benchmarks, metrics, and efficient algorithms for dynamic multiobjective optimization.

Purpose of the Study:

• To propose a novel dynamic MOEA capable of addressing the challenges of time-varying objectives.
• To enhance the convergence speed and tracking ability of MOEAs in dynamic environments.
• To improve the overall performance of evolutionary algorithms in dynamic multiobjective optimization.

Main Methods:

• Development of a new dynamic MOEA incorporating Kalman filter (KF) predictions in the decision space.
• Utilizing KF predictions to anticipate and guide the search towards evolving optima.
• Implementation of a scoring scheme to integrate KF predictions with a random reinitialization strategy.

Main Results:

• The proposed dynamic MOEA effectively tracks moving optima by leveraging predictive capabilities.
• KF predictions accelerate convergence towards the true optima in dynamic environments.
• Experimental comparisons show significant performance improvements over state-of-the-art dynamic MOEAs.

Conclusions:

• The KF-enhanced dynamic MOEA demonstrates superior performance in solving dynamic multiobjective optimization problems.
• Predictive modeling is a viable strategy for improving the adaptability of MOEAs to changing environments.
• The proposed approach offers a promising direction for future research in dynamic optimization.