A fast density peak clustering based particle swarm optimizer for dynamic optimization

Abstract

Dynamic optimization problems (DOPs) are optimization problems with time evolution characteristics. In this type of problem, the decision variables and the state variables change over time and produce results that will impact the future. To effectively address DOPs, this paper proposes a fast density peak clustering based particle swarm optimizer for dynamic optimization (DPCPSO). The main innovations of DPCPSO contain three critical components. First, a fast density peak clustering is applied to create multiple sub-populations, which can help the algorithm locate peaks. Second, stagnation detection is used to tackle the loss of diversity. Third, an optimal particle calibration strategy which can find the optimal solution quickly in a changing environment is proposed in response to environmental changes. Moreover, the hill climbing method is applied to help the memory quickly locate new peaks if the environment changes. The performance of our proposed algorithm has been tested on Mobile Peak Benchmark (MPB), Generalized Dynamic Benchmark Generator (GDBG) and Generalized Moving Peaks Benchmark (GMPB) problems and compared with seven state-of-the-art dynamic optimization algorithms. The experimental results validate the proposed algorithm performed competitively while solving DOPs.