高级认知粒子群优化模型及应用

Particle swarm optimization (PSO) is a swarm intelligent paradigm，inspired by the behavior of bird flocking, fish schooling and human society. At present, there exist drawbacks of premature convergence, serotinous convergence and misconvergence for PSO. In order to overcome the disadvantages of PSO, based on the analysis of cognition behavior characteristic of particles, this project aims to built the advance cognition behavior of each particle for improving the adaptivity of particles in complex environment, where the idea of non-social information processing models in modern cognitive psychology theory is used to explore the advance cognition behavior of particles. To design adaptive mechanism of information processing, perceptron module to environment, attention module, pattern recognition module and strategy decision module of particles are constructed. Then a PSO computing model with advance cognition is proposed and has a strong global searching ability in which particles adopt the adaptive mechanism of information processing. Meanwhile, the multi-scale learning mechanism is created and Lamarckian learning method is introduced to the new model. Copula theory, stochastic process theory, statistical analysis and other mathematical methods are employed to study the movement behavior of particles, convergence of algorithm, computational complexity of algorithm, and selection of learning parameters. Finally, to effectively address wireless sensor networks deployment, this project aims to apply the proposed model to solve this problem for global optimization. The systematic work obtained in this project from creative design of algorithm, theoretical analysis and application has the scientific value and broad application prospect in the field of engineering technology.

粒子群优化(Particle Swarm Optimization,PSO)算法是一种在鸟、鱼和人类社会行为规律启发下的群体智能范式。目前，PSO仍存在早熟和晚熟的收敛性缺陷。本项目拟在分析PSO行为特性的基础上，借鉴现代认知心理学理论，从非社会信息加工模型的新视角，构建粒子的高级认知行为以提高其在复杂动态环境下的自适应性。通过建立粒子的环境感知模块、注意模块、模式识别模块和策略决策模块来设计自适应的信息加工机制，同时引入拉马克学习方法和建立多尺度学习机制，提出具有全局收敛能力的高级认知PSO计算模型，并利用Copula、随机过程、统计分析等方法研究粒子的运动行为、算法收敛性、计算复杂性及学习参数选取等理论基础。将新模型应用于无线传感网络部署，为该类问题提供全局优化方法。本项目从算法的创新设计、理论分析及应用三个方面的系统性研究成果不仅具有科学价值，且在工程技术领域具有广阔的应用前景。

粒子群优化(Particle Swarm Optimization, PSO)算法在求解复杂问题时存在早熟和晚熟问题，针对该问题国内外学者已做了大量的研究。但是在保持算法简洁结构的前提下，同时避免算法的早熟和晚熟仍是PSO一个富有挑战性的问题。本项目首先分析了粒子的认知特性对种群多样性的影响，给出学习参数的概率分布与种群多样性之间的关系式；依据理论分析，提出了双学习模式PSO，该算法通过调整学习参数的概率特性控制群体的多样性。新算法不仅提高了优化性能同时保持了算法的简洁结构。接着发现骨干PSO收敛速度快但也易陷入早期收敛。对骨干PSO的运动行为做了理论分析，并获得了粒子的采样方差和群体中pbest的多样性与种群期望的关系式。依据理论分析，提出了并行协作骨干PSO，该算法在保持收敛精度的情况下大大加快了收敛速度。为了减少计算量和进一步增强群体的逃逸能力，提出了层次骨干PSO算法。最后将层次骨干PSO算法求解约束优化、动态优化和图像处理等问题，并取得良好的效果。