基于自学习迁移粒子群算法及高斯罚函数的无功优化方法  

Reactive Power Optimization Based on Self-Learning Migration Particle Swarm Optimization and Gaussian Penalty Function

作　　者：邓长虹[1] 马庆[1] 肖永 游佳斌[1] 李世春[1] 

机构地区：[1]武汉大学电气工程学院,湖北省武汉市430072 [2]贵州电力试验研究院,贵州省贵阳市550002

出　　处：《电网技术》2014年第12期3341-3346,共6页Power System Technology

基　　金：国家科技支撑计划(2013BAA02B02)~~

摘　　要：针对粒子群算法在求解无功优化问题时存在早熟收敛,易陷于局部最优的现象,提出了自学习迁移粒子群算法(self-learning migration particle swarm optimization,SLMPSO)。该算法在采用混沌序列对粒子群进行初始化操作,基于云模型理论的X-条件云发生器对粒子的惯性权重进行自适应调整的基础上,引入一种迁移操作,以引导全局最优粒子的飞行方向,解决粒子群后期朝单一进化方向进化的问题,有效地增强了算法的全局寻优能力。针对电力系统无功优化中的离散变量归整问题,首先将离散变量完全化为连续变量进行迭代求解,在寻求至全局最优解后引入高斯罚函数对离散变量进行归整操作。以网损和电压偏离最小为目标,对IEEE标准30节点算例进行仿真计算,验证了所提算法的有效性和可行性。In allusion to defects of particle swarm optimization （PSO） used to solve reactive power optimization problem, namely the premature convergence and easy to fall into local optimum, a self-learning migration particle swarm optimization （SLMPSO） algorithm is proposed. On the basis of adopting chaotic series to initialize particle swarm and performing adaptive adjustment of particles＇ inertial weights by cloud model theory based X-condition generator, a kind of migration operation is led in to lead the flight direction of global optimal particle to solve the problem that particles evolve towards single evolution direction in the later stage, thus the global searching ability of the proposed algorithm is enhanced effectively. To cope with the discrete variables in power system reactive power optimization, firstly, the discrete variables are entirely transformed into continuous variables to make the iterative solution, and after the global optimal solution is found the results of discretization operation based on Gaussian penalty function are rounded-off. Taking the minimum network loss and voltage deviation as objectives, the simulation of IEEE standard 30-bus system is performed, and the effectiveness and feasibility of the proposed algorithm are validated by simulation results.

关 键 词：云模型 迁移操作 粒子群优化算法 高斯罚函数 无功优化 

分 类 号：TM721[电气工程—电力系统及自动化]