ガウス過程回帰を用いたTwo-swarm Cooperative PSOの改良

抄録

Particle Swarm Optimization (PSO) is a type of evolutionary computation developed to mimic the behaviour of a flock of birds searching for food. Particles with positional information and velocity search for solutions as they move through the search space, sharing information across all particles to efficiently search for solutions. In PSO, only the positional information of the best solution is shared to update the velocity, which causes problems such as insufficient search, failure to find a global solution, and early convergence to a local solution. Sun and Li (2014) have proposed TCPSO (Two-swarm Cooperative Particle Swarm Optimization) with a slave particle swarm for intensive solution exploration. However, for high-dimensional and complex problems, even TCPSO sometimes falls into the trap of local solutions. This paper aims to improve the performance of TCPSO by using a Gaussian process to estimate the approximate shape of the function of the problem in the solution search process.