An Adaptive Multi-Objective Particle Swarm Optimization Method with Self-Learning Strategy

Multi-objective particle swarm optimization (MOPSO) algorithm is widely used in various engineering optimization problems. However, MOPSO has premature phenomenon and lacks the ability to balance convergence and diversity. To address these issues, several studies have proposed adaptive approaches focusing on parameter and leader selection. However, adopting multi-strategy switching as dynamic selection strategy exists redundant and inefficient operational characteristics. In addition, few studies have explored the learning capabilities of particles to dynamically adjust learning strategies based on evolutionary information. To tackle the aforementioned issues, this paper proposes a self-learning multi-objective particle swarm optimization (SLMOPSO) algorithm. Firstly, the proposed SLMOPSO selects leader candidates for population based on a parallel cell coordinate system (PCCS) in terms of convergence, diversity and distribution. Then, by utilizing a reinforcement learning algorithm with a trial-and-error mechanism, a three-dimensional Q-table is constructed for each particle. Finally, particles optimize and fine-tune their flight parameters and leaders adaptively according to their convergence and diversity states, so as to obtain maximum long-term profit. The superiority of the proposed SLMOPSO algorithm is verified by comparing it with three other different multi-objective optimization algorithms on seven test problems featuring multiple local optima and decentralized global optima.