Evolutionary algorithm is a hot topic in the researches on artificial intelligence. As an important research branch of evolutionary algorithm, particle swarm optimizer (PSO) has gained many important results on both theoretical analysis and engineering applications. However, PSO still has the problem of being trapped into local optima in solving high-dimensional and complex nonlinear problems. As one kind of stochastic algorithm, PSO has relatively lower searching efficiency than gradient based optimizers, which limits its common application in practical engineering problems. Recently, the study of PSO focuses on the performance improvement and its application in dynamic optimization and data mining, etc. The main contributions of this paper are summarized as follows:1. Based on the comprehensive learning PSO, the attractions of global and local best solutions are separately brought into each particle’s velocity updating to improve the algorithm’s convergence speed. Experimental results show these two improved algorithms can largely improve the performance of comprehensive learning PSO on most of the benchmarks.2. Combining with two improved comprehensive learning PSOs by an adaptive dynamic weighting strategy, an unified particle swarm optimizer based on dynamic neighborhood and comprehensive learning is proposed. Experimental results show that the proposed algorithm has faster search speed and higher precision compared with other state-of-the-art PSOs on most of the benchmarks.3. Dynamic optimization based on PSO is studied. A mapping function is used to improve the performance of PSO on dynamic optimization. The performance of various PSOs on dynamic optimization are tested. Combined with local optimizer, an adaptive PSO for dynamic optimization is proposed, in which a logistic regression classifier is trained to automatically determine the transition from PSO to local optimizer. Experiments on high-dimensional dynamic optimization problems show its effectiveness in solving these kinds of problems and the effectiveness in largely reducing the function evaluations. |