Fuzzy cognitive maps learning algorithm based on adaptive multi-factor particle swarm

Fuzzy Cognitive Maps (FCMs), as an effective soft computing method, has been widely applied in the modeling and analysis of complex systems. However, existing FCM learning algorithms do not fully consider the intricate causal relationships between nodes, making the learning process prone to local convergence.To address this issue, this paper proposes an Adaptive Multi-Factor Particle Swarm Optimization-Based Fuzzy Cognitive Map Learning Algorithm (AMFPSO-FCM). First, a decomposition strategy is employed to partition the FCM into multiple local links, treating the learning of each local link as an independent task, thereby modeling FCM learning as a multi-task learning problem. Subsequently, an FCM learning approach is designed based on the Multi-Factor Particle Swarm Optimization (MFPSO) algorithm, incorporating multi-dimensional dynamic chaotic inertia weights to enhance search capability and utilizing an adaptive probabilistic transition matrix to improve knowledge transfer efficiency. Experiments conducted on three air quality datasets demonstrate that AMFPSO-FCM outperforms existing algorithms in both accuracy and generalization ability.