Computing all global minima of an objective function is one difficulty research area. This paper proposes a hybrid approach combined with particle swarm optimization (PSO) and biomimetic pattern recognition (BPR), while PSO is used to detect the global minima and BPR is used to remember the previously detected global minima, as well as predicting the shape of the objective function. Furthermore, repulsion technique is applied to enhance the results. Experiment results show the proposed algorithm is effective and efficient.

Through mechanism analysis of particle swarm optimization (PSO), a new modified particle swarm optimization using differential equation group--differential evolutionary particle swarm optimization (DEPSO) is introduced, and the convergence is analyzed with translation function. To enhance the diversity of swarm and improve the global convergence of PSO, PID controller is used to control dynamic evolutionary behavior of DEPSO. Simulation results proved the algorithm’s efficiency.

The paper makes an analysis on the simulated mechanisms of mind evolutionary computation (MEC) firstly and proposes an extended computation model for MEC (EMEC). EMEC manipulates the search based on the behavior space and the information space. All operations in the behavior space are processed based on groups that symbolize the solution area, while the operations in the information space are done based on the billboards that are used to record the evolutionary information. All components of EMEC are formulated in details, including the similar-taxis operation, the cooperation operation, and a simulated-annealing -based dissimilation operation (SADO). EMEC emphasizes on the share and the guide of the information in the search, and gets a performance superior to the simple MEC. The proposed EMEC was performed on some well-known benchmark problems. The experimental results show EMEC is a robust global optimization algorithm and can alleviate the premature convergence validly.

The paper develops a self-organization particle swarm optimization (SOPSO) with the aim to alleviate the premature convergence. SOPSO emphasizes the information interactions between the particle-lever and the swarm-lever, and introduce feedback to simulate the function. Through the feedback information, the particles can perceive the swarm-lever state and adopt favorable behavior model to modify their behavior, which not only can modify the exploitation and the exploration of the algorithm adaptively, but also can vary the diversity of the swarm and contribute to a global optimum output in the swarm. Relative experiments have been done; the results show SOPSO performs very well on benchmark problems, and outperforms the basic PSO in search ability.

通过分析已有的几种微粒群算法，提出了一种统一模型，并通过线性控制理论分析了其收敛性能，同时利用李亚普诺夫函数给出了模型收敛速度的上界估计。为了进一步提高算法效率，提出了两种增强全局搜索性能的参数自适应算法：单群体参数自适应微粒群算法及双群体参数自适应微粒群算法。其中单群体参数自适应微粒群算法在进化初期使用算法发散的参数设置，从而能更大程度的提高算法全局收敛能力。双群体参数自适应微粒群算法使用两个种群，一个执行全局搜索，另一个执行局部搜索，通过信息交流以提高算法性能。仿真实例证明了算法的有效性。