An Escape from Local Optima Algorithm Based on Historical Feature Distribution for Solving Continuous Distributed Constraint Optimization Problems

Abstract

To address the issue that existing solution algorithms for Continuous Distributed Constraint Optimization Problems (C-DCOPs) are prone to getting trapped in local optima and failing to escape, this paper proposes a C-DCOP algorithm based on the historical feature distribution of agents (HFDA). In HFDA, agents' historical values are utilized to detect when the algorithm falls into a local optimum. Upon detection, the historical features of neighboring agents are constructed based on constraint functions. By leveraging the distribution of these features, a probabilistic constraint selection mechanism is introduced to explore new solutions that cannot be obtained under full constraints, thereby guiding the algorithm towards optimization and overcoming local optima or stagnation. The probabilistic neglect of neighbors based on their feature distribution increases the chances of finding better solutions while effectively suppressing noise. Comparative experiments on two benchmark problems demonstrate that HFDA achieves faster convergence and higher solution quality compared to state-of-the-art C-DCOP algorithms.