Learning to Guide Particle Search for Dynamic Multiobjective Optimization.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Dynamic multiobjective optimization problems (DMOPs) involve objectives that change over time, posing significant challenges for existing algorithms.
• Current dynamic multiobjective algorithms (DMOAs) struggle to learn and adapt to diverse environmental dynamics, limiting their effectiveness.
• Solving DMOPs efficiently online requires DMOAs with low computational overhead.

Purpose of the Study:

• To propose a novel particle search guidance network (PSGN) for effectively addressing DMOPs with various dynamics.
• To enable DMOAs to learn adaptive search strategies through reinforcement learning for improved performance in dynamic environments.
• To achieve computationally efficient solutions for DMOPs without compromising performance.

Main Methods:

• Development of a particle search guidance network (PSGN) to control individual search actions, including target selection and acceleration coefficients.
• Utilizing reinforcement learning to train the PSGN, allowing it to learn optimal actions for different environmental dynamics.
• Implementing incremental learning for efficient adjustment of PSGN hidden nodes and output weights to minimize computational cost.

Main Results:

• The proposed PSGN demonstrates capability in handling DMOPs across various dynamics.
• PSGN achieves computationally efficient guidance of particle search, suitable for online optimization tasks.
• Comparative experiments show PSGN outperforms seven state-of-the-art algorithms in solving DMOPs.

Conclusions:

• The PSGN effectively addresses the challenges of dynamic multiobjective optimization by learning adaptive search strategies.
• The proposed method offers a computationally efficient approach to solving DMOPs with diverse environmental changes.
• PSGN represents a significant advancement in DMOAs, providing robust and efficient solutions for complex dynamic optimization tasks.